Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
8f18cf13 | 19 | #include <linux/kernel.h> |
065631f6 | 20 | #include <linux/bio.h> |
39279cc3 | 21 | #include <linux/buffer_head.h> |
f2eb0a24 | 22 | #include <linux/file.h> |
39279cc3 CM |
23 | #include <linux/fs.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/highmem.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/init.h> | |
28 | #include <linux/string.h> | |
39279cc3 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/compat.h> | |
9ebefb18 | 35 | #include <linux/bit_spinlock.h> |
5103e947 | 36 | #include <linux/xattr.h> |
33268eaf | 37 | #include <linux/posix_acl.h> |
d899e052 | 38 | #include <linux/falloc.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
7a36ddec | 40 | #include <linux/ratelimit.h> |
22c44fe6 | 41 | #include <linux/mount.h> |
4b4e25f2 | 42 | #include "compat.h" |
39279cc3 CM |
43 | #include "ctree.h" |
44 | #include "disk-io.h" | |
45 | #include "transaction.h" | |
46 | #include "btrfs_inode.h" | |
47 | #include "ioctl.h" | |
48 | #include "print-tree.h" | |
e6dcd2dc | 49 | #include "ordered-data.h" |
95819c05 | 50 | #include "xattr.h" |
e02119d5 | 51 | #include "tree-log.h" |
4a54c8c1 | 52 | #include "volumes.h" |
c8b97818 | 53 | #include "compression.h" |
b4ce94de | 54 | #include "locking.h" |
dc89e982 | 55 | #include "free-space-cache.h" |
581bb050 | 56 | #include "inode-map.h" |
39279cc3 CM |
57 | |
58 | struct btrfs_iget_args { | |
59 | u64 ino; | |
60 | struct btrfs_root *root; | |
61 | }; | |
62 | ||
6e1d5dcc AD |
63 | static const struct inode_operations btrfs_dir_inode_operations; |
64 | static const struct inode_operations btrfs_symlink_inode_operations; | |
65 | static const struct inode_operations btrfs_dir_ro_inode_operations; | |
66 | static const struct inode_operations btrfs_special_inode_operations; | |
67 | static const struct inode_operations btrfs_file_inode_operations; | |
7f09410b AD |
68 | static const struct address_space_operations btrfs_aops; |
69 | static const struct address_space_operations btrfs_symlink_aops; | |
828c0950 | 70 | static const struct file_operations btrfs_dir_file_operations; |
d1310b2e | 71 | static struct extent_io_ops btrfs_extent_io_ops; |
39279cc3 CM |
72 | |
73 | static struct kmem_cache *btrfs_inode_cachep; | |
74 | struct kmem_cache *btrfs_trans_handle_cachep; | |
75 | struct kmem_cache *btrfs_transaction_cachep; | |
39279cc3 | 76 | struct kmem_cache *btrfs_path_cachep; |
dc89e982 | 77 | struct kmem_cache *btrfs_free_space_cachep; |
39279cc3 CM |
78 | |
79 | #define S_SHIFT 12 | |
80 | static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = { | |
81 | [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE, | |
82 | [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR, | |
83 | [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV, | |
84 | [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV, | |
85 | [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO, | |
86 | [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK, | |
87 | [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK, | |
88 | }; | |
89 | ||
a41ad394 JB |
90 | static int btrfs_setsize(struct inode *inode, loff_t newsize); |
91 | static int btrfs_truncate(struct inode *inode); | |
c8b97818 | 92 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end); |
771ed689 CM |
93 | static noinline int cow_file_range(struct inode *inode, |
94 | struct page *locked_page, | |
95 | u64 start, u64 end, int *page_started, | |
96 | unsigned long *nr_written, int unlock); | |
2115133f CM |
97 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, |
98 | struct btrfs_root *root, struct inode *inode); | |
7b128766 | 99 | |
f34f57a3 | 100 | static int btrfs_init_inode_security(struct btrfs_trans_handle *trans, |
2a7dba39 EP |
101 | struct inode *inode, struct inode *dir, |
102 | const struct qstr *qstr) | |
0279b4cd JO |
103 | { |
104 | int err; | |
105 | ||
f34f57a3 | 106 | err = btrfs_init_acl(trans, inode, dir); |
0279b4cd | 107 | if (!err) |
2a7dba39 | 108 | err = btrfs_xattr_security_init(trans, inode, dir, qstr); |
0279b4cd JO |
109 | return err; |
110 | } | |
111 | ||
c8b97818 CM |
112 | /* |
113 | * this does all the hard work for inserting an inline extent into | |
114 | * the btree. The caller should have done a btrfs_drop_extents so that | |
115 | * no overlapping inline items exist in the btree | |
116 | */ | |
d397712b | 117 | static noinline int insert_inline_extent(struct btrfs_trans_handle *trans, |
c8b97818 CM |
118 | struct btrfs_root *root, struct inode *inode, |
119 | u64 start, size_t size, size_t compressed_size, | |
fe3f566c | 120 | int compress_type, |
c8b97818 CM |
121 | struct page **compressed_pages) |
122 | { | |
123 | struct btrfs_key key; | |
124 | struct btrfs_path *path; | |
125 | struct extent_buffer *leaf; | |
126 | struct page *page = NULL; | |
127 | char *kaddr; | |
128 | unsigned long ptr; | |
129 | struct btrfs_file_extent_item *ei; | |
130 | int err = 0; | |
131 | int ret; | |
132 | size_t cur_size = size; | |
133 | size_t datasize; | |
134 | unsigned long offset; | |
c8b97818 | 135 | |
fe3f566c | 136 | if (compressed_size && compressed_pages) |
c8b97818 | 137 | cur_size = compressed_size; |
c8b97818 | 138 | |
d397712b CM |
139 | path = btrfs_alloc_path(); |
140 | if (!path) | |
c8b97818 CM |
141 | return -ENOMEM; |
142 | ||
b9473439 | 143 | path->leave_spinning = 1; |
c8b97818 | 144 | |
33345d01 | 145 | key.objectid = btrfs_ino(inode); |
c8b97818 CM |
146 | key.offset = start; |
147 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); | |
c8b97818 CM |
148 | datasize = btrfs_file_extent_calc_inline_size(cur_size); |
149 | ||
150 | inode_add_bytes(inode, size); | |
151 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
152 | datasize); | |
153 | BUG_ON(ret); | |
154 | if (ret) { | |
155 | err = ret; | |
c8b97818 CM |
156 | goto fail; |
157 | } | |
158 | leaf = path->nodes[0]; | |
159 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
160 | struct btrfs_file_extent_item); | |
161 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
162 | btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE); | |
163 | btrfs_set_file_extent_encryption(leaf, ei, 0); | |
164 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
165 | btrfs_set_file_extent_ram_bytes(leaf, ei, size); | |
166 | ptr = btrfs_file_extent_inline_start(ei); | |
167 | ||
261507a0 | 168 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 CM |
169 | struct page *cpage; |
170 | int i = 0; | |
d397712b | 171 | while (compressed_size > 0) { |
c8b97818 | 172 | cpage = compressed_pages[i]; |
5b050f04 | 173 | cur_size = min_t(unsigned long, compressed_size, |
c8b97818 CM |
174 | PAGE_CACHE_SIZE); |
175 | ||
b9473439 | 176 | kaddr = kmap_atomic(cpage, KM_USER0); |
c8b97818 | 177 | write_extent_buffer(leaf, kaddr, ptr, cur_size); |
b9473439 | 178 | kunmap_atomic(kaddr, KM_USER0); |
c8b97818 CM |
179 | |
180 | i++; | |
181 | ptr += cur_size; | |
182 | compressed_size -= cur_size; | |
183 | } | |
184 | btrfs_set_file_extent_compression(leaf, ei, | |
261507a0 | 185 | compress_type); |
c8b97818 CM |
186 | } else { |
187 | page = find_get_page(inode->i_mapping, | |
188 | start >> PAGE_CACHE_SHIFT); | |
189 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
190 | kaddr = kmap_atomic(page, KM_USER0); | |
191 | offset = start & (PAGE_CACHE_SIZE - 1); | |
192 | write_extent_buffer(leaf, kaddr + offset, ptr, size); | |
193 | kunmap_atomic(kaddr, KM_USER0); | |
194 | page_cache_release(page); | |
195 | } | |
196 | btrfs_mark_buffer_dirty(leaf); | |
197 | btrfs_free_path(path); | |
198 | ||
c2167754 YZ |
199 | /* |
200 | * we're an inline extent, so nobody can | |
201 | * extend the file past i_size without locking | |
202 | * a page we already have locked. | |
203 | * | |
204 | * We must do any isize and inode updates | |
205 | * before we unlock the pages. Otherwise we | |
206 | * could end up racing with unlink. | |
207 | */ | |
c8b97818 CM |
208 | BTRFS_I(inode)->disk_i_size = inode->i_size; |
209 | btrfs_update_inode(trans, root, inode); | |
c2167754 | 210 | |
c8b97818 CM |
211 | return 0; |
212 | fail: | |
213 | btrfs_free_path(path); | |
214 | return err; | |
215 | } | |
216 | ||
217 | ||
218 | /* | |
219 | * conditionally insert an inline extent into the file. This | |
220 | * does the checks required to make sure the data is small enough | |
221 | * to fit as an inline extent. | |
222 | */ | |
7f366cfe | 223 | static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans, |
c8b97818 CM |
224 | struct btrfs_root *root, |
225 | struct inode *inode, u64 start, u64 end, | |
fe3f566c | 226 | size_t compressed_size, int compress_type, |
c8b97818 CM |
227 | struct page **compressed_pages) |
228 | { | |
229 | u64 isize = i_size_read(inode); | |
230 | u64 actual_end = min(end + 1, isize); | |
231 | u64 inline_len = actual_end - start; | |
232 | u64 aligned_end = (end + root->sectorsize - 1) & | |
233 | ~((u64)root->sectorsize - 1); | |
234 | u64 hint_byte; | |
235 | u64 data_len = inline_len; | |
236 | int ret; | |
237 | ||
238 | if (compressed_size) | |
239 | data_len = compressed_size; | |
240 | ||
241 | if (start > 0 || | |
70b99e69 | 242 | actual_end >= PAGE_CACHE_SIZE || |
c8b97818 CM |
243 | data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) || |
244 | (!compressed_size && | |
245 | (actual_end & (root->sectorsize - 1)) == 0) || | |
246 | end + 1 < isize || | |
247 | data_len > root->fs_info->max_inline) { | |
248 | return 1; | |
249 | } | |
250 | ||
920bbbfb | 251 | ret = btrfs_drop_extents(trans, inode, start, aligned_end, |
a1ed835e | 252 | &hint_byte, 1); |
c8b97818 CM |
253 | BUG_ON(ret); |
254 | ||
255 | if (isize > actual_end) | |
256 | inline_len = min_t(u64, isize, actual_end); | |
257 | ret = insert_inline_extent(trans, root, inode, start, | |
258 | inline_len, compressed_size, | |
fe3f566c | 259 | compress_type, compressed_pages); |
c8b97818 | 260 | BUG_ON(ret); |
0ca1f7ce | 261 | btrfs_delalloc_release_metadata(inode, end + 1 - start); |
a1ed835e | 262 | btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0); |
c8b97818 CM |
263 | return 0; |
264 | } | |
265 | ||
771ed689 CM |
266 | struct async_extent { |
267 | u64 start; | |
268 | u64 ram_size; | |
269 | u64 compressed_size; | |
270 | struct page **pages; | |
271 | unsigned long nr_pages; | |
261507a0 | 272 | int compress_type; |
771ed689 CM |
273 | struct list_head list; |
274 | }; | |
275 | ||
276 | struct async_cow { | |
277 | struct inode *inode; | |
278 | struct btrfs_root *root; | |
279 | struct page *locked_page; | |
280 | u64 start; | |
281 | u64 end; | |
282 | struct list_head extents; | |
283 | struct btrfs_work work; | |
284 | }; | |
285 | ||
286 | static noinline int add_async_extent(struct async_cow *cow, | |
287 | u64 start, u64 ram_size, | |
288 | u64 compressed_size, | |
289 | struct page **pages, | |
261507a0 LZ |
290 | unsigned long nr_pages, |
291 | int compress_type) | |
771ed689 CM |
292 | { |
293 | struct async_extent *async_extent; | |
294 | ||
295 | async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS); | |
dac97e51 | 296 | BUG_ON(!async_extent); |
771ed689 CM |
297 | async_extent->start = start; |
298 | async_extent->ram_size = ram_size; | |
299 | async_extent->compressed_size = compressed_size; | |
300 | async_extent->pages = pages; | |
301 | async_extent->nr_pages = nr_pages; | |
261507a0 | 302 | async_extent->compress_type = compress_type; |
771ed689 CM |
303 | list_add_tail(&async_extent->list, &cow->extents); |
304 | return 0; | |
305 | } | |
306 | ||
d352ac68 | 307 | /* |
771ed689 CM |
308 | * we create compressed extents in two phases. The first |
309 | * phase compresses a range of pages that have already been | |
310 | * locked (both pages and state bits are locked). | |
c8b97818 | 311 | * |
771ed689 CM |
312 | * This is done inside an ordered work queue, and the compression |
313 | * is spread across many cpus. The actual IO submission is step | |
314 | * two, and the ordered work queue takes care of making sure that | |
315 | * happens in the same order things were put onto the queue by | |
316 | * writepages and friends. | |
c8b97818 | 317 | * |
771ed689 CM |
318 | * If this code finds it can't get good compression, it puts an |
319 | * entry onto the work queue to write the uncompressed bytes. This | |
320 | * makes sure that both compressed inodes and uncompressed inodes | |
321 | * are written in the same order that pdflush sent them down. | |
d352ac68 | 322 | */ |
771ed689 CM |
323 | static noinline int compress_file_range(struct inode *inode, |
324 | struct page *locked_page, | |
325 | u64 start, u64 end, | |
326 | struct async_cow *async_cow, | |
327 | int *num_added) | |
b888db2b CM |
328 | { |
329 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
330 | struct btrfs_trans_handle *trans; | |
db94535d | 331 | u64 num_bytes; |
db94535d | 332 | u64 blocksize = root->sectorsize; |
c8b97818 | 333 | u64 actual_end; |
42dc7bab | 334 | u64 isize = i_size_read(inode); |
e6dcd2dc | 335 | int ret = 0; |
c8b97818 CM |
336 | struct page **pages = NULL; |
337 | unsigned long nr_pages; | |
338 | unsigned long nr_pages_ret = 0; | |
339 | unsigned long total_compressed = 0; | |
340 | unsigned long total_in = 0; | |
341 | unsigned long max_compressed = 128 * 1024; | |
771ed689 | 342 | unsigned long max_uncompressed = 128 * 1024; |
c8b97818 CM |
343 | int i; |
344 | int will_compress; | |
261507a0 | 345 | int compress_type = root->fs_info->compress_type; |
b888db2b | 346 | |
4cb5300b CM |
347 | /* if this is a small write inside eof, kick off a defragbot */ |
348 | if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024) | |
349 | btrfs_add_inode_defrag(NULL, inode); | |
350 | ||
42dc7bab | 351 | actual_end = min_t(u64, isize, end + 1); |
c8b97818 CM |
352 | again: |
353 | will_compress = 0; | |
354 | nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1; | |
355 | nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE); | |
be20aa9d | 356 | |
f03d9301 CM |
357 | /* |
358 | * we don't want to send crud past the end of i_size through | |
359 | * compression, that's just a waste of CPU time. So, if the | |
360 | * end of the file is before the start of our current | |
361 | * requested range of bytes, we bail out to the uncompressed | |
362 | * cleanup code that can deal with all of this. | |
363 | * | |
364 | * It isn't really the fastest way to fix things, but this is a | |
365 | * very uncommon corner. | |
366 | */ | |
367 | if (actual_end <= start) | |
368 | goto cleanup_and_bail_uncompressed; | |
369 | ||
c8b97818 CM |
370 | total_compressed = actual_end - start; |
371 | ||
372 | /* we want to make sure that amount of ram required to uncompress | |
373 | * an extent is reasonable, so we limit the total size in ram | |
771ed689 CM |
374 | * of a compressed extent to 128k. This is a crucial number |
375 | * because it also controls how easily we can spread reads across | |
376 | * cpus for decompression. | |
377 | * | |
378 | * We also want to make sure the amount of IO required to do | |
379 | * a random read is reasonably small, so we limit the size of | |
380 | * a compressed extent to 128k. | |
c8b97818 CM |
381 | */ |
382 | total_compressed = min(total_compressed, max_uncompressed); | |
db94535d | 383 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
be20aa9d | 384 | num_bytes = max(blocksize, num_bytes); |
c8b97818 CM |
385 | total_in = 0; |
386 | ret = 0; | |
db94535d | 387 | |
771ed689 CM |
388 | /* |
389 | * we do compression for mount -o compress and when the | |
390 | * inode has not been flagged as nocompress. This flag can | |
391 | * change at any time if we discover bad compression ratios. | |
c8b97818 | 392 | */ |
6cbff00f | 393 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) && |
1e701a32 | 394 | (btrfs_test_opt(root, COMPRESS) || |
75e7cb7f LB |
395 | (BTRFS_I(inode)->force_compress) || |
396 | (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) { | |
c8b97818 | 397 | WARN_ON(pages); |
cfbc246e | 398 | pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS); |
560f7d75 LZ |
399 | if (!pages) { |
400 | /* just bail out to the uncompressed code */ | |
401 | goto cont; | |
402 | } | |
c8b97818 | 403 | |
261507a0 LZ |
404 | if (BTRFS_I(inode)->force_compress) |
405 | compress_type = BTRFS_I(inode)->force_compress; | |
406 | ||
407 | ret = btrfs_compress_pages(compress_type, | |
408 | inode->i_mapping, start, | |
409 | total_compressed, pages, | |
410 | nr_pages, &nr_pages_ret, | |
411 | &total_in, | |
412 | &total_compressed, | |
413 | max_compressed); | |
c8b97818 CM |
414 | |
415 | if (!ret) { | |
416 | unsigned long offset = total_compressed & | |
417 | (PAGE_CACHE_SIZE - 1); | |
418 | struct page *page = pages[nr_pages_ret - 1]; | |
419 | char *kaddr; | |
420 | ||
421 | /* zero the tail end of the last page, we might be | |
422 | * sending it down to disk | |
423 | */ | |
424 | if (offset) { | |
425 | kaddr = kmap_atomic(page, KM_USER0); | |
426 | memset(kaddr + offset, 0, | |
427 | PAGE_CACHE_SIZE - offset); | |
428 | kunmap_atomic(kaddr, KM_USER0); | |
429 | } | |
430 | will_compress = 1; | |
431 | } | |
432 | } | |
560f7d75 | 433 | cont: |
c8b97818 | 434 | if (start == 0) { |
7a7eaa40 | 435 | trans = btrfs_join_transaction(root); |
3612b495 | 436 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 437 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 438 | |
c8b97818 | 439 | /* lets try to make an inline extent */ |
771ed689 | 440 | if (ret || total_in < (actual_end - start)) { |
c8b97818 | 441 | /* we didn't compress the entire range, try |
771ed689 | 442 | * to make an uncompressed inline extent. |
c8b97818 CM |
443 | */ |
444 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 445 | start, end, 0, 0, NULL); |
c8b97818 | 446 | } else { |
771ed689 | 447 | /* try making a compressed inline extent */ |
c8b97818 CM |
448 | ret = cow_file_range_inline(trans, root, inode, |
449 | start, end, | |
fe3f566c LZ |
450 | total_compressed, |
451 | compress_type, pages); | |
c8b97818 CM |
452 | } |
453 | if (ret == 0) { | |
771ed689 CM |
454 | /* |
455 | * inline extent creation worked, we don't need | |
456 | * to create any more async work items. Unlock | |
457 | * and free up our temp pages. | |
458 | */ | |
c8b97818 | 459 | extent_clear_unlock_delalloc(inode, |
a791e35e CM |
460 | &BTRFS_I(inode)->io_tree, |
461 | start, end, NULL, | |
462 | EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY | | |
a3429ab7 | 463 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 464 | EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK); |
c2167754 YZ |
465 | |
466 | btrfs_end_transaction(trans, root); | |
c8b97818 CM |
467 | goto free_pages_out; |
468 | } | |
c2167754 | 469 | btrfs_end_transaction(trans, root); |
c8b97818 CM |
470 | } |
471 | ||
472 | if (will_compress) { | |
473 | /* | |
474 | * we aren't doing an inline extent round the compressed size | |
475 | * up to a block size boundary so the allocator does sane | |
476 | * things | |
477 | */ | |
478 | total_compressed = (total_compressed + blocksize - 1) & | |
479 | ~(blocksize - 1); | |
480 | ||
481 | /* | |
482 | * one last check to make sure the compression is really a | |
483 | * win, compare the page count read with the blocks on disk | |
484 | */ | |
485 | total_in = (total_in + PAGE_CACHE_SIZE - 1) & | |
486 | ~(PAGE_CACHE_SIZE - 1); | |
487 | if (total_compressed >= total_in) { | |
488 | will_compress = 0; | |
489 | } else { | |
c8b97818 CM |
490 | num_bytes = total_in; |
491 | } | |
492 | } | |
493 | if (!will_compress && pages) { | |
494 | /* | |
495 | * the compression code ran but failed to make things smaller, | |
496 | * free any pages it allocated and our page pointer array | |
497 | */ | |
498 | for (i = 0; i < nr_pages_ret; i++) { | |
70b99e69 | 499 | WARN_ON(pages[i]->mapping); |
c8b97818 CM |
500 | page_cache_release(pages[i]); |
501 | } | |
502 | kfree(pages); | |
503 | pages = NULL; | |
504 | total_compressed = 0; | |
505 | nr_pages_ret = 0; | |
506 | ||
507 | /* flag the file so we don't compress in the future */ | |
1e701a32 CM |
508 | if (!btrfs_test_opt(root, FORCE_COMPRESS) && |
509 | !(BTRFS_I(inode)->force_compress)) { | |
a555f810 | 510 | BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; |
1e701a32 | 511 | } |
c8b97818 | 512 | } |
771ed689 CM |
513 | if (will_compress) { |
514 | *num_added += 1; | |
c8b97818 | 515 | |
771ed689 CM |
516 | /* the async work queues will take care of doing actual |
517 | * allocation on disk for these compressed pages, | |
518 | * and will submit them to the elevator. | |
519 | */ | |
520 | add_async_extent(async_cow, start, num_bytes, | |
261507a0 LZ |
521 | total_compressed, pages, nr_pages_ret, |
522 | compress_type); | |
179e29e4 | 523 | |
24ae6365 | 524 | if (start + num_bytes < end) { |
771ed689 CM |
525 | start += num_bytes; |
526 | pages = NULL; | |
527 | cond_resched(); | |
528 | goto again; | |
529 | } | |
530 | } else { | |
f03d9301 | 531 | cleanup_and_bail_uncompressed: |
771ed689 CM |
532 | /* |
533 | * No compression, but we still need to write the pages in | |
534 | * the file we've been given so far. redirty the locked | |
535 | * page if it corresponds to our extent and set things up | |
536 | * for the async work queue to run cow_file_range to do | |
537 | * the normal delalloc dance | |
538 | */ | |
539 | if (page_offset(locked_page) >= start && | |
540 | page_offset(locked_page) <= end) { | |
541 | __set_page_dirty_nobuffers(locked_page); | |
542 | /* unlocked later on in the async handlers */ | |
543 | } | |
261507a0 LZ |
544 | add_async_extent(async_cow, start, end - start + 1, |
545 | 0, NULL, 0, BTRFS_COMPRESS_NONE); | |
771ed689 CM |
546 | *num_added += 1; |
547 | } | |
3b951516 | 548 | |
771ed689 CM |
549 | out: |
550 | return 0; | |
551 | ||
552 | free_pages_out: | |
553 | for (i = 0; i < nr_pages_ret; i++) { | |
554 | WARN_ON(pages[i]->mapping); | |
555 | page_cache_release(pages[i]); | |
556 | } | |
d397712b | 557 | kfree(pages); |
771ed689 CM |
558 | |
559 | goto out; | |
560 | } | |
561 | ||
562 | /* | |
563 | * phase two of compressed writeback. This is the ordered portion | |
564 | * of the code, which only gets called in the order the work was | |
565 | * queued. We walk all the async extents created by compress_file_range | |
566 | * and send them down to the disk. | |
567 | */ | |
568 | static noinline int submit_compressed_extents(struct inode *inode, | |
569 | struct async_cow *async_cow) | |
570 | { | |
571 | struct async_extent *async_extent; | |
572 | u64 alloc_hint = 0; | |
573 | struct btrfs_trans_handle *trans; | |
574 | struct btrfs_key ins; | |
575 | struct extent_map *em; | |
576 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
577 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
578 | struct extent_io_tree *io_tree; | |
f5a84ee3 | 579 | int ret = 0; |
771ed689 CM |
580 | |
581 | if (list_empty(&async_cow->extents)) | |
582 | return 0; | |
583 | ||
771ed689 | 584 | |
d397712b | 585 | while (!list_empty(&async_cow->extents)) { |
771ed689 CM |
586 | async_extent = list_entry(async_cow->extents.next, |
587 | struct async_extent, list); | |
588 | list_del(&async_extent->list); | |
c8b97818 | 589 | |
771ed689 CM |
590 | io_tree = &BTRFS_I(inode)->io_tree; |
591 | ||
f5a84ee3 | 592 | retry: |
771ed689 CM |
593 | /* did the compression code fall back to uncompressed IO? */ |
594 | if (!async_extent->pages) { | |
595 | int page_started = 0; | |
596 | unsigned long nr_written = 0; | |
597 | ||
598 | lock_extent(io_tree, async_extent->start, | |
2ac55d41 JB |
599 | async_extent->start + |
600 | async_extent->ram_size - 1, GFP_NOFS); | |
771ed689 CM |
601 | |
602 | /* allocate blocks */ | |
f5a84ee3 JB |
603 | ret = cow_file_range(inode, async_cow->locked_page, |
604 | async_extent->start, | |
605 | async_extent->start + | |
606 | async_extent->ram_size - 1, | |
607 | &page_started, &nr_written, 0); | |
771ed689 CM |
608 | |
609 | /* | |
610 | * if page_started, cow_file_range inserted an | |
611 | * inline extent and took care of all the unlocking | |
612 | * and IO for us. Otherwise, we need to submit | |
613 | * all those pages down to the drive. | |
614 | */ | |
f5a84ee3 | 615 | if (!page_started && !ret) |
771ed689 CM |
616 | extent_write_locked_range(io_tree, |
617 | inode, async_extent->start, | |
d397712b | 618 | async_extent->start + |
771ed689 CM |
619 | async_extent->ram_size - 1, |
620 | btrfs_get_extent, | |
621 | WB_SYNC_ALL); | |
622 | kfree(async_extent); | |
623 | cond_resched(); | |
624 | continue; | |
625 | } | |
626 | ||
627 | lock_extent(io_tree, async_extent->start, | |
628 | async_extent->start + async_extent->ram_size - 1, | |
629 | GFP_NOFS); | |
771ed689 | 630 | |
7a7eaa40 | 631 | trans = btrfs_join_transaction(root); |
3612b495 | 632 | BUG_ON(IS_ERR(trans)); |
74b21075 | 633 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 CM |
634 | ret = btrfs_reserve_extent(trans, root, |
635 | async_extent->compressed_size, | |
636 | async_extent->compressed_size, | |
637 | 0, alloc_hint, | |
638 | (u64)-1, &ins, 1); | |
c2167754 YZ |
639 | btrfs_end_transaction(trans, root); |
640 | ||
f5a84ee3 JB |
641 | if (ret) { |
642 | int i; | |
643 | for (i = 0; i < async_extent->nr_pages; i++) { | |
644 | WARN_ON(async_extent->pages[i]->mapping); | |
645 | page_cache_release(async_extent->pages[i]); | |
646 | } | |
647 | kfree(async_extent->pages); | |
648 | async_extent->nr_pages = 0; | |
649 | async_extent->pages = NULL; | |
650 | unlock_extent(io_tree, async_extent->start, | |
651 | async_extent->start + | |
652 | async_extent->ram_size - 1, GFP_NOFS); | |
653 | goto retry; | |
654 | } | |
655 | ||
c2167754 YZ |
656 | /* |
657 | * here we're doing allocation and writeback of the | |
658 | * compressed pages | |
659 | */ | |
660 | btrfs_drop_extent_cache(inode, async_extent->start, | |
661 | async_extent->start + | |
662 | async_extent->ram_size - 1, 0); | |
663 | ||
172ddd60 | 664 | em = alloc_extent_map(); |
c26a9203 | 665 | BUG_ON(!em); |
771ed689 CM |
666 | em->start = async_extent->start; |
667 | em->len = async_extent->ram_size; | |
445a6944 | 668 | em->orig_start = em->start; |
c8b97818 | 669 | |
771ed689 CM |
670 | em->block_start = ins.objectid; |
671 | em->block_len = ins.offset; | |
672 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
261507a0 | 673 | em->compress_type = async_extent->compress_type; |
771ed689 CM |
674 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
675 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); | |
676 | ||
d397712b | 677 | while (1) { |
890871be | 678 | write_lock(&em_tree->lock); |
771ed689 | 679 | ret = add_extent_mapping(em_tree, em); |
890871be | 680 | write_unlock(&em_tree->lock); |
771ed689 CM |
681 | if (ret != -EEXIST) { |
682 | free_extent_map(em); | |
683 | break; | |
684 | } | |
685 | btrfs_drop_extent_cache(inode, async_extent->start, | |
686 | async_extent->start + | |
687 | async_extent->ram_size - 1, 0); | |
688 | } | |
689 | ||
261507a0 LZ |
690 | ret = btrfs_add_ordered_extent_compress(inode, |
691 | async_extent->start, | |
692 | ins.objectid, | |
693 | async_extent->ram_size, | |
694 | ins.offset, | |
695 | BTRFS_ORDERED_COMPRESSED, | |
696 | async_extent->compress_type); | |
771ed689 CM |
697 | BUG_ON(ret); |
698 | ||
771ed689 CM |
699 | /* |
700 | * clear dirty, set writeback and unlock the pages. | |
701 | */ | |
702 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
703 | &BTRFS_I(inode)->io_tree, |
704 | async_extent->start, | |
705 | async_extent->start + | |
706 | async_extent->ram_size - 1, | |
707 | NULL, EXTENT_CLEAR_UNLOCK_PAGE | | |
708 | EXTENT_CLEAR_UNLOCK | | |
a3429ab7 | 709 | EXTENT_CLEAR_DELALLOC | |
a791e35e | 710 | EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK); |
771ed689 CM |
711 | |
712 | ret = btrfs_submit_compressed_write(inode, | |
d397712b CM |
713 | async_extent->start, |
714 | async_extent->ram_size, | |
715 | ins.objectid, | |
716 | ins.offset, async_extent->pages, | |
717 | async_extent->nr_pages); | |
771ed689 CM |
718 | |
719 | BUG_ON(ret); | |
771ed689 CM |
720 | alloc_hint = ins.objectid + ins.offset; |
721 | kfree(async_extent); | |
722 | cond_resched(); | |
723 | } | |
724 | ||
771ed689 CM |
725 | return 0; |
726 | } | |
727 | ||
4b46fce2 JB |
728 | static u64 get_extent_allocation_hint(struct inode *inode, u64 start, |
729 | u64 num_bytes) | |
730 | { | |
731 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
732 | struct extent_map *em; | |
733 | u64 alloc_hint = 0; | |
734 | ||
735 | read_lock(&em_tree->lock); | |
736 | em = search_extent_mapping(em_tree, start, num_bytes); | |
737 | if (em) { | |
738 | /* | |
739 | * if block start isn't an actual block number then find the | |
740 | * first block in this inode and use that as a hint. If that | |
741 | * block is also bogus then just don't worry about it. | |
742 | */ | |
743 | if (em->block_start >= EXTENT_MAP_LAST_BYTE) { | |
744 | free_extent_map(em); | |
745 | em = search_extent_mapping(em_tree, 0, 0); | |
746 | if (em && em->block_start < EXTENT_MAP_LAST_BYTE) | |
747 | alloc_hint = em->block_start; | |
748 | if (em) | |
749 | free_extent_map(em); | |
750 | } else { | |
751 | alloc_hint = em->block_start; | |
752 | free_extent_map(em); | |
753 | } | |
754 | } | |
755 | read_unlock(&em_tree->lock); | |
756 | ||
757 | return alloc_hint; | |
758 | } | |
759 | ||
771ed689 CM |
760 | /* |
761 | * when extent_io.c finds a delayed allocation range in the file, | |
762 | * the call backs end up in this code. The basic idea is to | |
763 | * allocate extents on disk for the range, and create ordered data structs | |
764 | * in ram to track those extents. | |
765 | * | |
766 | * locked_page is the page that writepage had locked already. We use | |
767 | * it to make sure we don't do extra locks or unlocks. | |
768 | * | |
769 | * *page_started is set to one if we unlock locked_page and do everything | |
770 | * required to start IO on it. It may be clean and already done with | |
771 | * IO when we return. | |
772 | */ | |
773 | static noinline int cow_file_range(struct inode *inode, | |
774 | struct page *locked_page, | |
775 | u64 start, u64 end, int *page_started, | |
776 | unsigned long *nr_written, | |
777 | int unlock) | |
778 | { | |
779 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
780 | struct btrfs_trans_handle *trans; | |
781 | u64 alloc_hint = 0; | |
782 | u64 num_bytes; | |
783 | unsigned long ram_size; | |
784 | u64 disk_num_bytes; | |
785 | u64 cur_alloc_size; | |
786 | u64 blocksize = root->sectorsize; | |
771ed689 CM |
787 | struct btrfs_key ins; |
788 | struct extent_map *em; | |
789 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
790 | int ret = 0; | |
791 | ||
2cf8572d | 792 | BUG_ON(btrfs_is_free_space_inode(root, inode)); |
7a7eaa40 | 793 | trans = btrfs_join_transaction(root); |
3612b495 | 794 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 795 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
771ed689 | 796 | |
771ed689 CM |
797 | num_bytes = (end - start + blocksize) & ~(blocksize - 1); |
798 | num_bytes = max(blocksize, num_bytes); | |
799 | disk_num_bytes = num_bytes; | |
800 | ret = 0; | |
801 | ||
4cb5300b CM |
802 | /* if this is a small write inside eof, kick off defrag */ |
803 | if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024) | |
804 | btrfs_add_inode_defrag(trans, inode); | |
805 | ||
771ed689 CM |
806 | if (start == 0) { |
807 | /* lets try to make an inline extent */ | |
808 | ret = cow_file_range_inline(trans, root, inode, | |
fe3f566c | 809 | start, end, 0, 0, NULL); |
771ed689 CM |
810 | if (ret == 0) { |
811 | extent_clear_unlock_delalloc(inode, | |
a791e35e CM |
812 | &BTRFS_I(inode)->io_tree, |
813 | start, end, NULL, | |
814 | EXTENT_CLEAR_UNLOCK_PAGE | | |
815 | EXTENT_CLEAR_UNLOCK | | |
816 | EXTENT_CLEAR_DELALLOC | | |
817 | EXTENT_CLEAR_DIRTY | | |
818 | EXTENT_SET_WRITEBACK | | |
819 | EXTENT_END_WRITEBACK); | |
c2167754 | 820 | |
771ed689 CM |
821 | *nr_written = *nr_written + |
822 | (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE; | |
823 | *page_started = 1; | |
824 | ret = 0; | |
825 | goto out; | |
826 | } | |
827 | } | |
828 | ||
829 | BUG_ON(disk_num_bytes > | |
6c41761f | 830 | btrfs_super_total_bytes(root->fs_info->super_copy)); |
771ed689 | 831 | |
4b46fce2 | 832 | alloc_hint = get_extent_allocation_hint(inode, start, num_bytes); |
771ed689 CM |
833 | btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0); |
834 | ||
d397712b | 835 | while (disk_num_bytes > 0) { |
a791e35e CM |
836 | unsigned long op; |
837 | ||
287a0ab9 | 838 | cur_alloc_size = disk_num_bytes; |
e6dcd2dc | 839 | ret = btrfs_reserve_extent(trans, root, cur_alloc_size, |
771ed689 | 840 | root->sectorsize, 0, alloc_hint, |
e6dcd2dc | 841 | (u64)-1, &ins, 1); |
d397712b CM |
842 | BUG_ON(ret); |
843 | ||
172ddd60 | 844 | em = alloc_extent_map(); |
c26a9203 | 845 | BUG_ON(!em); |
e6dcd2dc | 846 | em->start = start; |
445a6944 | 847 | em->orig_start = em->start; |
771ed689 CM |
848 | ram_size = ins.offset; |
849 | em->len = ins.offset; | |
c8b97818 | 850 | |
e6dcd2dc | 851 | em->block_start = ins.objectid; |
c8b97818 | 852 | em->block_len = ins.offset; |
e6dcd2dc | 853 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
7f3c74fb | 854 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
c8b97818 | 855 | |
d397712b | 856 | while (1) { |
890871be | 857 | write_lock(&em_tree->lock); |
e6dcd2dc | 858 | ret = add_extent_mapping(em_tree, em); |
890871be | 859 | write_unlock(&em_tree->lock); |
e6dcd2dc CM |
860 | if (ret != -EEXIST) { |
861 | free_extent_map(em); | |
862 | break; | |
863 | } | |
864 | btrfs_drop_extent_cache(inode, start, | |
c8b97818 | 865 | start + ram_size - 1, 0); |
e6dcd2dc CM |
866 | } |
867 | ||
98d20f67 | 868 | cur_alloc_size = ins.offset; |
e6dcd2dc | 869 | ret = btrfs_add_ordered_extent(inode, start, ins.objectid, |
771ed689 | 870 | ram_size, cur_alloc_size, 0); |
e6dcd2dc | 871 | BUG_ON(ret); |
c8b97818 | 872 | |
17d217fe YZ |
873 | if (root->root_key.objectid == |
874 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
875 | ret = btrfs_reloc_clone_csums(inode, start, | |
876 | cur_alloc_size); | |
877 | BUG_ON(ret); | |
878 | } | |
879 | ||
d397712b | 880 | if (disk_num_bytes < cur_alloc_size) |
3b951516 | 881 | break; |
d397712b | 882 | |
c8b97818 CM |
883 | /* we're not doing compressed IO, don't unlock the first |
884 | * page (which the caller expects to stay locked), don't | |
885 | * clear any dirty bits and don't set any writeback bits | |
8b62b72b CM |
886 | * |
887 | * Do set the Private2 bit so we know this page was properly | |
888 | * setup for writepage | |
c8b97818 | 889 | */ |
a791e35e CM |
890 | op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0; |
891 | op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
892 | EXTENT_SET_PRIVATE2; | |
893 | ||
c8b97818 CM |
894 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
895 | start, start + ram_size - 1, | |
a791e35e | 896 | locked_page, op); |
c8b97818 | 897 | disk_num_bytes -= cur_alloc_size; |
c59f8951 CM |
898 | num_bytes -= cur_alloc_size; |
899 | alloc_hint = ins.objectid + ins.offset; | |
900 | start += cur_alloc_size; | |
b888db2b | 901 | } |
b888db2b | 902 | out: |
771ed689 | 903 | ret = 0; |
b888db2b | 904 | btrfs_end_transaction(trans, root); |
c8b97818 | 905 | |
be20aa9d | 906 | return ret; |
771ed689 | 907 | } |
c8b97818 | 908 | |
771ed689 CM |
909 | /* |
910 | * work queue call back to started compression on a file and pages | |
911 | */ | |
912 | static noinline void async_cow_start(struct btrfs_work *work) | |
913 | { | |
914 | struct async_cow *async_cow; | |
915 | int num_added = 0; | |
916 | async_cow = container_of(work, struct async_cow, work); | |
917 | ||
918 | compress_file_range(async_cow->inode, async_cow->locked_page, | |
919 | async_cow->start, async_cow->end, async_cow, | |
920 | &num_added); | |
921 | if (num_added == 0) | |
922 | async_cow->inode = NULL; | |
923 | } | |
924 | ||
925 | /* | |
926 | * work queue call back to submit previously compressed pages | |
927 | */ | |
928 | static noinline void async_cow_submit(struct btrfs_work *work) | |
929 | { | |
930 | struct async_cow *async_cow; | |
931 | struct btrfs_root *root; | |
932 | unsigned long nr_pages; | |
933 | ||
934 | async_cow = container_of(work, struct async_cow, work); | |
935 | ||
936 | root = async_cow->root; | |
937 | nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >> | |
938 | PAGE_CACHE_SHIFT; | |
939 | ||
940 | atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages); | |
941 | ||
942 | if (atomic_read(&root->fs_info->async_delalloc_pages) < | |
943 | 5 * 1042 * 1024 && | |
944 | waitqueue_active(&root->fs_info->async_submit_wait)) | |
945 | wake_up(&root->fs_info->async_submit_wait); | |
946 | ||
d397712b | 947 | if (async_cow->inode) |
771ed689 | 948 | submit_compressed_extents(async_cow->inode, async_cow); |
771ed689 | 949 | } |
c8b97818 | 950 | |
771ed689 CM |
951 | static noinline void async_cow_free(struct btrfs_work *work) |
952 | { | |
953 | struct async_cow *async_cow; | |
954 | async_cow = container_of(work, struct async_cow, work); | |
955 | kfree(async_cow); | |
956 | } | |
957 | ||
958 | static int cow_file_range_async(struct inode *inode, struct page *locked_page, | |
959 | u64 start, u64 end, int *page_started, | |
960 | unsigned long *nr_written) | |
961 | { | |
962 | struct async_cow *async_cow; | |
963 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
964 | unsigned long nr_pages; | |
965 | u64 cur_end; | |
966 | int limit = 10 * 1024 * 1042; | |
967 | ||
a3429ab7 CM |
968 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED, |
969 | 1, 0, NULL, GFP_NOFS); | |
d397712b | 970 | while (start < end) { |
771ed689 | 971 | async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS); |
8d413713 | 972 | BUG_ON(!async_cow); |
771ed689 CM |
973 | async_cow->inode = inode; |
974 | async_cow->root = root; | |
975 | async_cow->locked_page = locked_page; | |
976 | async_cow->start = start; | |
977 | ||
6cbff00f | 978 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) |
771ed689 CM |
979 | cur_end = end; |
980 | else | |
981 | cur_end = min(end, start + 512 * 1024 - 1); | |
982 | ||
983 | async_cow->end = cur_end; | |
984 | INIT_LIST_HEAD(&async_cow->extents); | |
985 | ||
986 | async_cow->work.func = async_cow_start; | |
987 | async_cow->work.ordered_func = async_cow_submit; | |
988 | async_cow->work.ordered_free = async_cow_free; | |
989 | async_cow->work.flags = 0; | |
990 | ||
771ed689 CM |
991 | nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >> |
992 | PAGE_CACHE_SHIFT; | |
993 | atomic_add(nr_pages, &root->fs_info->async_delalloc_pages); | |
994 | ||
995 | btrfs_queue_worker(&root->fs_info->delalloc_workers, | |
996 | &async_cow->work); | |
997 | ||
998 | if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) { | |
999 | wait_event(root->fs_info->async_submit_wait, | |
1000 | (atomic_read(&root->fs_info->async_delalloc_pages) < | |
1001 | limit)); | |
1002 | } | |
1003 | ||
d397712b | 1004 | while (atomic_read(&root->fs_info->async_submit_draining) && |
771ed689 CM |
1005 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
1006 | wait_event(root->fs_info->async_submit_wait, | |
1007 | (atomic_read(&root->fs_info->async_delalloc_pages) == | |
1008 | 0)); | |
1009 | } | |
1010 | ||
1011 | *nr_written += nr_pages; | |
1012 | start = cur_end + 1; | |
1013 | } | |
1014 | *page_started = 1; | |
1015 | return 0; | |
be20aa9d CM |
1016 | } |
1017 | ||
d397712b | 1018 | static noinline int csum_exist_in_range(struct btrfs_root *root, |
17d217fe YZ |
1019 | u64 bytenr, u64 num_bytes) |
1020 | { | |
1021 | int ret; | |
1022 | struct btrfs_ordered_sum *sums; | |
1023 | LIST_HEAD(list); | |
1024 | ||
07d400a6 | 1025 | ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr, |
a2de733c | 1026 | bytenr + num_bytes - 1, &list, 0); |
17d217fe YZ |
1027 | if (ret == 0 && list_empty(&list)) |
1028 | return 0; | |
1029 | ||
1030 | while (!list_empty(&list)) { | |
1031 | sums = list_entry(list.next, struct btrfs_ordered_sum, list); | |
1032 | list_del(&sums->list); | |
1033 | kfree(sums); | |
1034 | } | |
1035 | return 1; | |
1036 | } | |
1037 | ||
d352ac68 CM |
1038 | /* |
1039 | * when nowcow writeback call back. This checks for snapshots or COW copies | |
1040 | * of the extents that exist in the file, and COWs the file as required. | |
1041 | * | |
1042 | * If no cow copies or snapshots exist, we write directly to the existing | |
1043 | * blocks on disk | |
1044 | */ | |
7f366cfe CM |
1045 | static noinline int run_delalloc_nocow(struct inode *inode, |
1046 | struct page *locked_page, | |
771ed689 CM |
1047 | u64 start, u64 end, int *page_started, int force, |
1048 | unsigned long *nr_written) | |
be20aa9d | 1049 | { |
be20aa9d | 1050 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7ea394f1 | 1051 | struct btrfs_trans_handle *trans; |
be20aa9d | 1052 | struct extent_buffer *leaf; |
be20aa9d | 1053 | struct btrfs_path *path; |
80ff3856 | 1054 | struct btrfs_file_extent_item *fi; |
be20aa9d | 1055 | struct btrfs_key found_key; |
80ff3856 YZ |
1056 | u64 cow_start; |
1057 | u64 cur_offset; | |
1058 | u64 extent_end; | |
5d4f98a2 | 1059 | u64 extent_offset; |
80ff3856 YZ |
1060 | u64 disk_bytenr; |
1061 | u64 num_bytes; | |
1062 | int extent_type; | |
1063 | int ret; | |
d899e052 | 1064 | int type; |
80ff3856 YZ |
1065 | int nocow; |
1066 | int check_prev = 1; | |
82d5902d | 1067 | bool nolock; |
33345d01 | 1068 | u64 ino = btrfs_ino(inode); |
be20aa9d CM |
1069 | |
1070 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1071 | if (!path) |
1072 | return -ENOMEM; | |
82d5902d | 1073 | |
2cf8572d | 1074 | nolock = btrfs_is_free_space_inode(root, inode); |
82d5902d LZ |
1075 | |
1076 | if (nolock) | |
7a7eaa40 | 1077 | trans = btrfs_join_transaction_nolock(root); |
82d5902d | 1078 | else |
7a7eaa40 | 1079 | trans = btrfs_join_transaction(root); |
ff5714cc | 1080 | |
3612b495 | 1081 | BUG_ON(IS_ERR(trans)); |
74b21075 | 1082 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
be20aa9d | 1083 | |
80ff3856 YZ |
1084 | cow_start = (u64)-1; |
1085 | cur_offset = start; | |
1086 | while (1) { | |
33345d01 | 1087 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
80ff3856 YZ |
1088 | cur_offset, 0); |
1089 | BUG_ON(ret < 0); | |
1090 | if (ret > 0 && path->slots[0] > 0 && check_prev) { | |
1091 | leaf = path->nodes[0]; | |
1092 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1093 | path->slots[0] - 1); | |
33345d01 | 1094 | if (found_key.objectid == ino && |
80ff3856 YZ |
1095 | found_key.type == BTRFS_EXTENT_DATA_KEY) |
1096 | path->slots[0]--; | |
1097 | } | |
1098 | check_prev = 0; | |
1099 | next_slot: | |
1100 | leaf = path->nodes[0]; | |
1101 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1102 | ret = btrfs_next_leaf(root, path); | |
1103 | if (ret < 0) | |
1104 | BUG_ON(1); | |
1105 | if (ret > 0) | |
1106 | break; | |
1107 | leaf = path->nodes[0]; | |
1108 | } | |
be20aa9d | 1109 | |
80ff3856 YZ |
1110 | nocow = 0; |
1111 | disk_bytenr = 0; | |
17d217fe | 1112 | num_bytes = 0; |
80ff3856 YZ |
1113 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
1114 | ||
33345d01 | 1115 | if (found_key.objectid > ino || |
80ff3856 YZ |
1116 | found_key.type > BTRFS_EXTENT_DATA_KEY || |
1117 | found_key.offset > end) | |
1118 | break; | |
1119 | ||
1120 | if (found_key.offset > cur_offset) { | |
1121 | extent_end = found_key.offset; | |
e9061e21 | 1122 | extent_type = 0; |
80ff3856 YZ |
1123 | goto out_check; |
1124 | } | |
1125 | ||
1126 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1127 | struct btrfs_file_extent_item); | |
1128 | extent_type = btrfs_file_extent_type(leaf, fi); | |
1129 | ||
d899e052 YZ |
1130 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
1131 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
80ff3856 | 1132 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
5d4f98a2 | 1133 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
80ff3856 YZ |
1134 | extent_end = found_key.offset + |
1135 | btrfs_file_extent_num_bytes(leaf, fi); | |
1136 | if (extent_end <= start) { | |
1137 | path->slots[0]++; | |
1138 | goto next_slot; | |
1139 | } | |
17d217fe YZ |
1140 | if (disk_bytenr == 0) |
1141 | goto out_check; | |
80ff3856 YZ |
1142 | if (btrfs_file_extent_compression(leaf, fi) || |
1143 | btrfs_file_extent_encryption(leaf, fi) || | |
1144 | btrfs_file_extent_other_encoding(leaf, fi)) | |
1145 | goto out_check; | |
d899e052 YZ |
1146 | if (extent_type == BTRFS_FILE_EXTENT_REG && !force) |
1147 | goto out_check; | |
d2fb3437 | 1148 | if (btrfs_extent_readonly(root, disk_bytenr)) |
80ff3856 | 1149 | goto out_check; |
33345d01 | 1150 | if (btrfs_cross_ref_exist(trans, root, ino, |
5d4f98a2 YZ |
1151 | found_key.offset - |
1152 | extent_offset, disk_bytenr)) | |
17d217fe | 1153 | goto out_check; |
5d4f98a2 | 1154 | disk_bytenr += extent_offset; |
17d217fe YZ |
1155 | disk_bytenr += cur_offset - found_key.offset; |
1156 | num_bytes = min(end + 1, extent_end) - cur_offset; | |
1157 | /* | |
1158 | * force cow if csum exists in the range. | |
1159 | * this ensure that csum for a given extent are | |
1160 | * either valid or do not exist. | |
1161 | */ | |
1162 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
1163 | goto out_check; | |
80ff3856 YZ |
1164 | nocow = 1; |
1165 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
1166 | extent_end = found_key.offset + | |
1167 | btrfs_file_extent_inline_len(leaf, fi); | |
1168 | extent_end = ALIGN(extent_end, root->sectorsize); | |
1169 | } else { | |
1170 | BUG_ON(1); | |
1171 | } | |
1172 | out_check: | |
1173 | if (extent_end <= start) { | |
1174 | path->slots[0]++; | |
1175 | goto next_slot; | |
1176 | } | |
1177 | if (!nocow) { | |
1178 | if (cow_start == (u64)-1) | |
1179 | cow_start = cur_offset; | |
1180 | cur_offset = extent_end; | |
1181 | if (cur_offset > end) | |
1182 | break; | |
1183 | path->slots[0]++; | |
1184 | goto next_slot; | |
7ea394f1 YZ |
1185 | } |
1186 | ||
b3b4aa74 | 1187 | btrfs_release_path(path); |
80ff3856 YZ |
1188 | if (cow_start != (u64)-1) { |
1189 | ret = cow_file_range(inode, locked_page, cow_start, | |
771ed689 CM |
1190 | found_key.offset - 1, page_started, |
1191 | nr_written, 1); | |
80ff3856 YZ |
1192 | BUG_ON(ret); |
1193 | cow_start = (u64)-1; | |
7ea394f1 | 1194 | } |
80ff3856 | 1195 | |
d899e052 YZ |
1196 | if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
1197 | struct extent_map *em; | |
1198 | struct extent_map_tree *em_tree; | |
1199 | em_tree = &BTRFS_I(inode)->extent_tree; | |
172ddd60 | 1200 | em = alloc_extent_map(); |
c26a9203 | 1201 | BUG_ON(!em); |
d899e052 | 1202 | em->start = cur_offset; |
445a6944 | 1203 | em->orig_start = em->start; |
d899e052 YZ |
1204 | em->len = num_bytes; |
1205 | em->block_len = num_bytes; | |
1206 | em->block_start = disk_bytenr; | |
1207 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
1208 | set_bit(EXTENT_FLAG_PINNED, &em->flags); | |
1209 | while (1) { | |
890871be | 1210 | write_lock(&em_tree->lock); |
d899e052 | 1211 | ret = add_extent_mapping(em_tree, em); |
890871be | 1212 | write_unlock(&em_tree->lock); |
d899e052 YZ |
1213 | if (ret != -EEXIST) { |
1214 | free_extent_map(em); | |
1215 | break; | |
1216 | } | |
1217 | btrfs_drop_extent_cache(inode, em->start, | |
1218 | em->start + em->len - 1, 0); | |
1219 | } | |
1220 | type = BTRFS_ORDERED_PREALLOC; | |
1221 | } else { | |
1222 | type = BTRFS_ORDERED_NOCOW; | |
1223 | } | |
80ff3856 YZ |
1224 | |
1225 | ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr, | |
d899e052 YZ |
1226 | num_bytes, num_bytes, type); |
1227 | BUG_ON(ret); | |
771ed689 | 1228 | |
efa56464 YZ |
1229 | if (root->root_key.objectid == |
1230 | BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1231 | ret = btrfs_reloc_clone_csums(inode, cur_offset, | |
1232 | num_bytes); | |
1233 | BUG_ON(ret); | |
1234 | } | |
1235 | ||
d899e052 | 1236 | extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree, |
a791e35e CM |
1237 | cur_offset, cur_offset + num_bytes - 1, |
1238 | locked_page, EXTENT_CLEAR_UNLOCK_PAGE | | |
1239 | EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC | | |
1240 | EXTENT_SET_PRIVATE2); | |
80ff3856 YZ |
1241 | cur_offset = extent_end; |
1242 | if (cur_offset > end) | |
1243 | break; | |
be20aa9d | 1244 | } |
b3b4aa74 | 1245 | btrfs_release_path(path); |
80ff3856 YZ |
1246 | |
1247 | if (cur_offset <= end && cow_start == (u64)-1) | |
1248 | cow_start = cur_offset; | |
1249 | if (cow_start != (u64)-1) { | |
1250 | ret = cow_file_range(inode, locked_page, cow_start, end, | |
771ed689 | 1251 | page_started, nr_written, 1); |
80ff3856 YZ |
1252 | BUG_ON(ret); |
1253 | } | |
1254 | ||
0cb59c99 JB |
1255 | if (nolock) { |
1256 | ret = btrfs_end_transaction_nolock(trans, root); | |
1257 | BUG_ON(ret); | |
1258 | } else { | |
1259 | ret = btrfs_end_transaction(trans, root); | |
1260 | BUG_ON(ret); | |
1261 | } | |
7ea394f1 | 1262 | btrfs_free_path(path); |
80ff3856 | 1263 | return 0; |
be20aa9d CM |
1264 | } |
1265 | ||
d352ac68 CM |
1266 | /* |
1267 | * extent_io.c call back to do delayed allocation processing | |
1268 | */ | |
c8b97818 | 1269 | static int run_delalloc_range(struct inode *inode, struct page *locked_page, |
771ed689 CM |
1270 | u64 start, u64 end, int *page_started, |
1271 | unsigned long *nr_written) | |
be20aa9d | 1272 | { |
be20aa9d | 1273 | int ret; |
7f366cfe | 1274 | struct btrfs_root *root = BTRFS_I(inode)->root; |
a2135011 | 1275 | |
6cbff00f | 1276 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) |
c8b97818 | 1277 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1278 | page_started, 1, nr_written); |
6cbff00f | 1279 | else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) |
d899e052 | 1280 | ret = run_delalloc_nocow(inode, locked_page, start, end, |
d397712b | 1281 | page_started, 0, nr_written); |
1e701a32 | 1282 | else if (!btrfs_test_opt(root, COMPRESS) && |
75e7cb7f LB |
1283 | !(BTRFS_I(inode)->force_compress) && |
1284 | !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) | |
7f366cfe CM |
1285 | ret = cow_file_range(inode, locked_page, start, end, |
1286 | page_started, nr_written, 1); | |
be20aa9d | 1287 | else |
771ed689 | 1288 | ret = cow_file_range_async(inode, locked_page, start, end, |
d397712b | 1289 | page_started, nr_written); |
b888db2b CM |
1290 | return ret; |
1291 | } | |
1292 | ||
1bf85046 JM |
1293 | static void btrfs_split_extent_hook(struct inode *inode, |
1294 | struct extent_state *orig, u64 split) | |
9ed74f2d | 1295 | { |
0ca1f7ce | 1296 | /* not delalloc, ignore it */ |
9ed74f2d | 1297 | if (!(orig->state & EXTENT_DELALLOC)) |
1bf85046 | 1298 | return; |
9ed74f2d | 1299 | |
9e0baf60 JB |
1300 | spin_lock(&BTRFS_I(inode)->lock); |
1301 | BTRFS_I(inode)->outstanding_extents++; | |
1302 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1303 | } |
1304 | ||
1305 | /* | |
1306 | * extent_io.c merge_extent_hook, used to track merged delayed allocation | |
1307 | * extents so we can keep track of new extents that are just merged onto old | |
1308 | * extents, such as when we are doing sequential writes, so we can properly | |
1309 | * account for the metadata space we'll need. | |
1310 | */ | |
1bf85046 JM |
1311 | static void btrfs_merge_extent_hook(struct inode *inode, |
1312 | struct extent_state *new, | |
1313 | struct extent_state *other) | |
9ed74f2d | 1314 | { |
9ed74f2d JB |
1315 | /* not delalloc, ignore it */ |
1316 | if (!(other->state & EXTENT_DELALLOC)) | |
1bf85046 | 1317 | return; |
9ed74f2d | 1318 | |
9e0baf60 JB |
1319 | spin_lock(&BTRFS_I(inode)->lock); |
1320 | BTRFS_I(inode)->outstanding_extents--; | |
1321 | spin_unlock(&BTRFS_I(inode)->lock); | |
9ed74f2d JB |
1322 | } |
1323 | ||
d352ac68 CM |
1324 | /* |
1325 | * extent_io.c set_bit_hook, used to track delayed allocation | |
1326 | * bytes in this file, and to maintain the list of inodes that | |
1327 | * have pending delalloc work to be done. | |
1328 | */ | |
1bf85046 JM |
1329 | static void btrfs_set_bit_hook(struct inode *inode, |
1330 | struct extent_state *state, int *bits) | |
291d673e | 1331 | { |
9ed74f2d | 1332 | |
75eff68e CM |
1333 | /* |
1334 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1335 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1336 | * bit, which is only set or cleared with irqs on |
1337 | */ | |
0ca1f7ce | 1338 | if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1339 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1340 | u64 len = state->end + 1 - state->start; |
2cf8572d | 1341 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
9ed74f2d | 1342 | |
9e0baf60 | 1343 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1344 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1345 | } else { |
1346 | spin_lock(&BTRFS_I(inode)->lock); | |
1347 | BTRFS_I(inode)->outstanding_extents++; | |
1348 | spin_unlock(&BTRFS_I(inode)->lock); | |
1349 | } | |
287a0ab9 | 1350 | |
75eff68e | 1351 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1352 | BTRFS_I(inode)->delalloc_bytes += len; |
1353 | root->fs_info->delalloc_bytes += len; | |
0cb59c99 | 1354 | if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
ea8c2819 CM |
1355 | list_add_tail(&BTRFS_I(inode)->delalloc_inodes, |
1356 | &root->fs_info->delalloc_inodes); | |
1357 | } | |
75eff68e | 1358 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1359 | } |
291d673e CM |
1360 | } |
1361 | ||
d352ac68 CM |
1362 | /* |
1363 | * extent_io.c clear_bit_hook, see set_bit_hook for why | |
1364 | */ | |
1bf85046 JM |
1365 | static void btrfs_clear_bit_hook(struct inode *inode, |
1366 | struct extent_state *state, int *bits) | |
291d673e | 1367 | { |
75eff68e CM |
1368 | /* |
1369 | * set_bit and clear bit hooks normally require _irqsave/restore | |
27160b6b | 1370 | * but in this case, we are only testing for the DELALLOC |
75eff68e CM |
1371 | * bit, which is only set or cleared with irqs on |
1372 | */ | |
0ca1f7ce | 1373 | if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) { |
291d673e | 1374 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1375 | u64 len = state->end + 1 - state->start; |
2cf8572d | 1376 | bool do_list = !btrfs_is_free_space_inode(root, inode); |
bcbfce8a | 1377 | |
9e0baf60 | 1378 | if (*bits & EXTENT_FIRST_DELALLOC) { |
0ca1f7ce | 1379 | *bits &= ~EXTENT_FIRST_DELALLOC; |
9e0baf60 JB |
1380 | } else if (!(*bits & EXTENT_DO_ACCOUNTING)) { |
1381 | spin_lock(&BTRFS_I(inode)->lock); | |
1382 | BTRFS_I(inode)->outstanding_extents--; | |
1383 | spin_unlock(&BTRFS_I(inode)->lock); | |
1384 | } | |
0ca1f7ce YZ |
1385 | |
1386 | if (*bits & EXTENT_DO_ACCOUNTING) | |
1387 | btrfs_delalloc_release_metadata(inode, len); | |
1388 | ||
0cb59c99 JB |
1389 | if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID |
1390 | && do_list) | |
0ca1f7ce | 1391 | btrfs_free_reserved_data_space(inode, len); |
9ed74f2d | 1392 | |
75eff68e | 1393 | spin_lock(&root->fs_info->delalloc_lock); |
0ca1f7ce YZ |
1394 | root->fs_info->delalloc_bytes -= len; |
1395 | BTRFS_I(inode)->delalloc_bytes -= len; | |
1396 | ||
0cb59c99 | 1397 | if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 && |
ea8c2819 CM |
1398 | !list_empty(&BTRFS_I(inode)->delalloc_inodes)) { |
1399 | list_del_init(&BTRFS_I(inode)->delalloc_inodes); | |
1400 | } | |
75eff68e | 1401 | spin_unlock(&root->fs_info->delalloc_lock); |
291d673e | 1402 | } |
291d673e CM |
1403 | } |
1404 | ||
d352ac68 CM |
1405 | /* |
1406 | * extent_io.c merge_bio_hook, this must check the chunk tree to make sure | |
1407 | * we don't create bios that span stripes or chunks | |
1408 | */ | |
239b14b3 | 1409 | int btrfs_merge_bio_hook(struct page *page, unsigned long offset, |
c8b97818 CM |
1410 | size_t size, struct bio *bio, |
1411 | unsigned long bio_flags) | |
239b14b3 CM |
1412 | { |
1413 | struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; | |
1414 | struct btrfs_mapping_tree *map_tree; | |
a62b9401 | 1415 | u64 logical = (u64)bio->bi_sector << 9; |
239b14b3 CM |
1416 | u64 length = 0; |
1417 | u64 map_length; | |
239b14b3 CM |
1418 | int ret; |
1419 | ||
771ed689 CM |
1420 | if (bio_flags & EXTENT_BIO_COMPRESSED) |
1421 | return 0; | |
1422 | ||
f2d8d74d | 1423 | length = bio->bi_size; |
239b14b3 CM |
1424 | map_tree = &root->fs_info->mapping_tree; |
1425 | map_length = length; | |
cea9e445 | 1426 | ret = btrfs_map_block(map_tree, READ, logical, |
f188591e | 1427 | &map_length, NULL, 0); |
cea9e445 | 1428 | |
d397712b | 1429 | if (map_length < length + size) |
239b14b3 | 1430 | return 1; |
411fc6bc | 1431 | return ret; |
239b14b3 CM |
1432 | } |
1433 | ||
d352ac68 CM |
1434 | /* |
1435 | * in order to insert checksums into the metadata in large chunks, | |
1436 | * we wait until bio submission time. All the pages in the bio are | |
1437 | * checksummed and sums are attached onto the ordered extent record. | |
1438 | * | |
1439 | * At IO completion time the cums attached on the ordered extent record | |
1440 | * are inserted into the btree | |
1441 | */ | |
d397712b CM |
1442 | static int __btrfs_submit_bio_start(struct inode *inode, int rw, |
1443 | struct bio *bio, int mirror_num, | |
eaf25d93 CM |
1444 | unsigned long bio_flags, |
1445 | u64 bio_offset) | |
065631f6 | 1446 | { |
065631f6 | 1447 | struct btrfs_root *root = BTRFS_I(inode)->root; |
065631f6 | 1448 | int ret = 0; |
e015640f | 1449 | |
d20f7043 | 1450 | ret = btrfs_csum_one_bio(root, inode, bio, 0, 0); |
44b8bd7e | 1451 | BUG_ON(ret); |
4a69a410 CM |
1452 | return 0; |
1453 | } | |
e015640f | 1454 | |
4a69a410 CM |
1455 | /* |
1456 | * in order to insert checksums into the metadata in large chunks, | |
1457 | * we wait until bio submission time. All the pages in the bio are | |
1458 | * checksummed and sums are attached onto the ordered extent record. | |
1459 | * | |
1460 | * At IO completion time the cums attached on the ordered extent record | |
1461 | * are inserted into the btree | |
1462 | */ | |
b2950863 | 1463 | static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1464 | int mirror_num, unsigned long bio_flags, |
1465 | u64 bio_offset) | |
4a69a410 CM |
1466 | { |
1467 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
8b712842 | 1468 | return btrfs_map_bio(root, rw, bio, mirror_num, 1); |
44b8bd7e CM |
1469 | } |
1470 | ||
d352ac68 | 1471 | /* |
cad321ad CM |
1472 | * extent_io.c submission hook. This does the right thing for csum calculation |
1473 | * on write, or reading the csums from the tree before a read | |
d352ac68 | 1474 | */ |
b2950863 | 1475 | static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, |
eaf25d93 CM |
1476 | int mirror_num, unsigned long bio_flags, |
1477 | u64 bio_offset) | |
44b8bd7e CM |
1478 | { |
1479 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1480 | int ret = 0; | |
19b9bdb0 | 1481 | int skip_sum; |
44b8bd7e | 1482 | |
6cbff00f | 1483 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
cad321ad | 1484 | |
2cf8572d | 1485 | if (btrfs_is_free_space_inode(root, inode)) |
0cb59c99 JB |
1486 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2); |
1487 | else | |
1488 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
e6dcd2dc | 1489 | BUG_ON(ret); |
065631f6 | 1490 | |
7b6d91da | 1491 | if (!(rw & REQ_WRITE)) { |
d20f7043 | 1492 | if (bio_flags & EXTENT_BIO_COMPRESSED) { |
c8b97818 CM |
1493 | return btrfs_submit_compressed_read(inode, bio, |
1494 | mirror_num, bio_flags); | |
c2db1073 TI |
1495 | } else if (!skip_sum) { |
1496 | ret = btrfs_lookup_bio_sums(root, inode, bio, NULL); | |
1497 | if (ret) | |
1498 | return ret; | |
1499 | } | |
4d1b5fb4 | 1500 | goto mapit; |
19b9bdb0 | 1501 | } else if (!skip_sum) { |
17d217fe YZ |
1502 | /* csum items have already been cloned */ |
1503 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) | |
1504 | goto mapit; | |
19b9bdb0 CM |
1505 | /* we're doing a write, do the async checksumming */ |
1506 | return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, | |
44b8bd7e | 1507 | inode, rw, bio, mirror_num, |
eaf25d93 CM |
1508 | bio_flags, bio_offset, |
1509 | __btrfs_submit_bio_start, | |
4a69a410 | 1510 | __btrfs_submit_bio_done); |
19b9bdb0 CM |
1511 | } |
1512 | ||
0b86a832 | 1513 | mapit: |
8b712842 | 1514 | return btrfs_map_bio(root, rw, bio, mirror_num, 0); |
065631f6 | 1515 | } |
6885f308 | 1516 | |
d352ac68 CM |
1517 | /* |
1518 | * given a list of ordered sums record them in the inode. This happens | |
1519 | * at IO completion time based on sums calculated at bio submission time. | |
1520 | */ | |
ba1da2f4 | 1521 | static noinline int add_pending_csums(struct btrfs_trans_handle *trans, |
e6dcd2dc CM |
1522 | struct inode *inode, u64 file_offset, |
1523 | struct list_head *list) | |
1524 | { | |
e6dcd2dc CM |
1525 | struct btrfs_ordered_sum *sum; |
1526 | ||
c6e30871 | 1527 | list_for_each_entry(sum, list, list) { |
d20f7043 CM |
1528 | btrfs_csum_file_blocks(trans, |
1529 | BTRFS_I(inode)->root->fs_info->csum_root, sum); | |
e6dcd2dc CM |
1530 | } |
1531 | return 0; | |
1532 | } | |
1533 | ||
2ac55d41 JB |
1534 | int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end, |
1535 | struct extent_state **cached_state) | |
ea8c2819 | 1536 | { |
d397712b | 1537 | if ((end & (PAGE_CACHE_SIZE - 1)) == 0) |
771ed689 | 1538 | WARN_ON(1); |
ea8c2819 | 1539 | return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end, |
2ac55d41 | 1540 | cached_state, GFP_NOFS); |
ea8c2819 CM |
1541 | } |
1542 | ||
d352ac68 | 1543 | /* see btrfs_writepage_start_hook for details on why this is required */ |
247e743c CM |
1544 | struct btrfs_writepage_fixup { |
1545 | struct page *page; | |
1546 | struct btrfs_work work; | |
1547 | }; | |
1548 | ||
b2950863 | 1549 | static void btrfs_writepage_fixup_worker(struct btrfs_work *work) |
247e743c CM |
1550 | { |
1551 | struct btrfs_writepage_fixup *fixup; | |
1552 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 1553 | struct extent_state *cached_state = NULL; |
247e743c CM |
1554 | struct page *page; |
1555 | struct inode *inode; | |
1556 | u64 page_start; | |
1557 | u64 page_end; | |
87826df0 | 1558 | int ret; |
247e743c CM |
1559 | |
1560 | fixup = container_of(work, struct btrfs_writepage_fixup, work); | |
1561 | page = fixup->page; | |
4a096752 | 1562 | again: |
247e743c CM |
1563 | lock_page(page); |
1564 | if (!page->mapping || !PageDirty(page) || !PageChecked(page)) { | |
1565 | ClearPageChecked(page); | |
1566 | goto out_page; | |
1567 | } | |
1568 | ||
1569 | inode = page->mapping->host; | |
1570 | page_start = page_offset(page); | |
1571 | page_end = page_offset(page) + PAGE_CACHE_SIZE - 1; | |
1572 | ||
2ac55d41 JB |
1573 | lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0, |
1574 | &cached_state, GFP_NOFS); | |
4a096752 CM |
1575 | |
1576 | /* already ordered? We're done */ | |
8b62b72b | 1577 | if (PagePrivate2(page)) |
247e743c | 1578 | goto out; |
4a096752 CM |
1579 | |
1580 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
1581 | if (ordered) { | |
2ac55d41 JB |
1582 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, |
1583 | page_end, &cached_state, GFP_NOFS); | |
4a096752 CM |
1584 | unlock_page(page); |
1585 | btrfs_start_ordered_extent(inode, ordered, 1); | |
87826df0 | 1586 | btrfs_put_ordered_extent(ordered); |
4a096752 CM |
1587 | goto again; |
1588 | } | |
247e743c | 1589 | |
87826df0 JM |
1590 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
1591 | if (ret) { | |
1592 | mapping_set_error(page->mapping, ret); | |
1593 | end_extent_writepage(page, ret, page_start, page_end); | |
1594 | ClearPageChecked(page); | |
1595 | goto out; | |
1596 | } | |
1597 | ||
2ac55d41 | 1598 | btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state); |
247e743c | 1599 | ClearPageChecked(page); |
87826df0 | 1600 | set_page_dirty(page); |
247e743c | 1601 | out: |
2ac55d41 JB |
1602 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end, |
1603 | &cached_state, GFP_NOFS); | |
247e743c CM |
1604 | out_page: |
1605 | unlock_page(page); | |
1606 | page_cache_release(page); | |
b897abec | 1607 | kfree(fixup); |
247e743c CM |
1608 | } |
1609 | ||
1610 | /* | |
1611 | * There are a few paths in the higher layers of the kernel that directly | |
1612 | * set the page dirty bit without asking the filesystem if it is a | |
1613 | * good idea. This causes problems because we want to make sure COW | |
1614 | * properly happens and the data=ordered rules are followed. | |
1615 | * | |
c8b97818 | 1616 | * In our case any range that doesn't have the ORDERED bit set |
247e743c CM |
1617 | * hasn't been properly setup for IO. We kick off an async process |
1618 | * to fix it up. The async helper will wait for ordered extents, set | |
1619 | * the delalloc bit and make it safe to write the page. | |
1620 | */ | |
b2950863 | 1621 | static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end) |
247e743c CM |
1622 | { |
1623 | struct inode *inode = page->mapping->host; | |
1624 | struct btrfs_writepage_fixup *fixup; | |
1625 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
247e743c | 1626 | |
8b62b72b CM |
1627 | /* this page is properly in the ordered list */ |
1628 | if (TestClearPagePrivate2(page)) | |
247e743c CM |
1629 | return 0; |
1630 | ||
1631 | if (PageChecked(page)) | |
1632 | return -EAGAIN; | |
1633 | ||
1634 | fixup = kzalloc(sizeof(*fixup), GFP_NOFS); | |
1635 | if (!fixup) | |
1636 | return -EAGAIN; | |
f421950f | 1637 | |
247e743c CM |
1638 | SetPageChecked(page); |
1639 | page_cache_get(page); | |
1640 | fixup->work.func = btrfs_writepage_fixup_worker; | |
1641 | fixup->page = page; | |
1642 | btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work); | |
87826df0 | 1643 | return -EBUSY; |
247e743c CM |
1644 | } |
1645 | ||
d899e052 YZ |
1646 | static int insert_reserved_file_extent(struct btrfs_trans_handle *trans, |
1647 | struct inode *inode, u64 file_pos, | |
1648 | u64 disk_bytenr, u64 disk_num_bytes, | |
1649 | u64 num_bytes, u64 ram_bytes, | |
1650 | u8 compression, u8 encryption, | |
1651 | u16 other_encoding, int extent_type) | |
1652 | { | |
1653 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
1654 | struct btrfs_file_extent_item *fi; | |
1655 | struct btrfs_path *path; | |
1656 | struct extent_buffer *leaf; | |
1657 | struct btrfs_key ins; | |
1658 | u64 hint; | |
1659 | int ret; | |
1660 | ||
1661 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
1662 | if (!path) |
1663 | return -ENOMEM; | |
d899e052 | 1664 | |
b9473439 | 1665 | path->leave_spinning = 1; |
a1ed835e CM |
1666 | |
1667 | /* | |
1668 | * we may be replacing one extent in the tree with another. | |
1669 | * The new extent is pinned in the extent map, and we don't want | |
1670 | * to drop it from the cache until it is completely in the btree. | |
1671 | * | |
1672 | * So, tell btrfs_drop_extents to leave this extent in the cache. | |
1673 | * the caller is expected to unpin it and allow it to be merged | |
1674 | * with the others. | |
1675 | */ | |
920bbbfb YZ |
1676 | ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes, |
1677 | &hint, 0); | |
d899e052 YZ |
1678 | BUG_ON(ret); |
1679 | ||
33345d01 | 1680 | ins.objectid = btrfs_ino(inode); |
d899e052 YZ |
1681 | ins.offset = file_pos; |
1682 | ins.type = BTRFS_EXTENT_DATA_KEY; | |
1683 | ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi)); | |
1684 | BUG_ON(ret); | |
1685 | leaf = path->nodes[0]; | |
1686 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
1687 | struct btrfs_file_extent_item); | |
1688 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); | |
1689 | btrfs_set_file_extent_type(leaf, fi, extent_type); | |
1690 | btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr); | |
1691 | btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes); | |
1692 | btrfs_set_file_extent_offset(leaf, fi, 0); | |
1693 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); | |
1694 | btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes); | |
1695 | btrfs_set_file_extent_compression(leaf, fi, compression); | |
1696 | btrfs_set_file_extent_encryption(leaf, fi, encryption); | |
1697 | btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); | |
b9473439 CM |
1698 | |
1699 | btrfs_unlock_up_safe(path, 1); | |
1700 | btrfs_set_lock_blocking(leaf); | |
1701 | ||
d899e052 YZ |
1702 | btrfs_mark_buffer_dirty(leaf); |
1703 | ||
1704 | inode_add_bytes(inode, num_bytes); | |
d899e052 YZ |
1705 | |
1706 | ins.objectid = disk_bytenr; | |
1707 | ins.offset = disk_num_bytes; | |
1708 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
5d4f98a2 YZ |
1709 | ret = btrfs_alloc_reserved_file_extent(trans, root, |
1710 | root->root_key.objectid, | |
33345d01 | 1711 | btrfs_ino(inode), file_pos, &ins); |
d899e052 | 1712 | BUG_ON(ret); |
d899e052 | 1713 | btrfs_free_path(path); |
b9473439 | 1714 | |
d899e052 YZ |
1715 | return 0; |
1716 | } | |
1717 | ||
5d13a98f CM |
1718 | /* |
1719 | * helper function for btrfs_finish_ordered_io, this | |
1720 | * just reads in some of the csum leaves to prime them into ram | |
1721 | * before we start the transaction. It limits the amount of btree | |
1722 | * reads required while inside the transaction. | |
1723 | */ | |
d352ac68 CM |
1724 | /* as ordered data IO finishes, this gets called so we can finish |
1725 | * an ordered extent if the range of bytes in the file it covers are | |
1726 | * fully written. | |
1727 | */ | |
211f90e6 | 1728 | static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end) |
e6dcd2dc | 1729 | { |
e6dcd2dc | 1730 | struct btrfs_root *root = BTRFS_I(inode)->root; |
0ca1f7ce | 1731 | struct btrfs_trans_handle *trans = NULL; |
5d13a98f | 1732 | struct btrfs_ordered_extent *ordered_extent = NULL; |
e6dcd2dc | 1733 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
2ac55d41 | 1734 | struct extent_state *cached_state = NULL; |
261507a0 | 1735 | int compress_type = 0; |
e6dcd2dc | 1736 | int ret; |
82d5902d | 1737 | bool nolock; |
e6dcd2dc | 1738 | |
5a1a3df1 JB |
1739 | ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start, |
1740 | end - start + 1); | |
ba1da2f4 | 1741 | if (!ret) |
e6dcd2dc | 1742 | return 0; |
e6dcd2dc | 1743 | BUG_ON(!ordered_extent); |
efd049fb | 1744 | |
2cf8572d | 1745 | nolock = btrfs_is_free_space_inode(root, inode); |
0cb59c99 | 1746 | |
c2167754 YZ |
1747 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) { |
1748 | BUG_ON(!list_empty(&ordered_extent->list)); | |
1749 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); | |
1750 | if (!ret) { | |
0cb59c99 | 1751 | if (nolock) |
7a7eaa40 | 1752 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1753 | else |
7a7eaa40 | 1754 | trans = btrfs_join_transaction(root); |
3612b495 | 1755 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 1756 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
2115133f | 1757 | ret = btrfs_update_inode_fallback(trans, root, inode); |
c2167754 | 1758 | BUG_ON(ret); |
c2167754 YZ |
1759 | } |
1760 | goto out; | |
1761 | } | |
e6dcd2dc | 1762 | |
2ac55d41 JB |
1763 | lock_extent_bits(io_tree, ordered_extent->file_offset, |
1764 | ordered_extent->file_offset + ordered_extent->len - 1, | |
1765 | 0, &cached_state, GFP_NOFS); | |
e6dcd2dc | 1766 | |
0cb59c99 | 1767 | if (nolock) |
7a7eaa40 | 1768 | trans = btrfs_join_transaction_nolock(root); |
0cb59c99 | 1769 | else |
7a7eaa40 | 1770 | trans = btrfs_join_transaction(root); |
3612b495 | 1771 | BUG_ON(IS_ERR(trans)); |
0ca1f7ce | 1772 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
c2167754 | 1773 | |
c8b97818 | 1774 | if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags)) |
261507a0 | 1775 | compress_type = ordered_extent->compress_type; |
d899e052 | 1776 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
261507a0 | 1777 | BUG_ON(compress_type); |
920bbbfb | 1778 | ret = btrfs_mark_extent_written(trans, inode, |
d899e052 YZ |
1779 | ordered_extent->file_offset, |
1780 | ordered_extent->file_offset + | |
1781 | ordered_extent->len); | |
1782 | BUG_ON(ret); | |
1783 | } else { | |
0af3d00b | 1784 | BUG_ON(root == root->fs_info->tree_root); |
d899e052 YZ |
1785 | ret = insert_reserved_file_extent(trans, inode, |
1786 | ordered_extent->file_offset, | |
1787 | ordered_extent->start, | |
1788 | ordered_extent->disk_len, | |
1789 | ordered_extent->len, | |
1790 | ordered_extent->len, | |
261507a0 | 1791 | compress_type, 0, 0, |
d899e052 | 1792 | BTRFS_FILE_EXTENT_REG); |
a1ed835e CM |
1793 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, |
1794 | ordered_extent->file_offset, | |
1795 | ordered_extent->len); | |
d899e052 YZ |
1796 | BUG_ON(ret); |
1797 | } | |
2ac55d41 JB |
1798 | unlock_extent_cached(io_tree, ordered_extent->file_offset, |
1799 | ordered_extent->file_offset + | |
1800 | ordered_extent->len - 1, &cached_state, GFP_NOFS); | |
1801 | ||
e6dcd2dc CM |
1802 | add_pending_csums(trans, inode, ordered_extent->file_offset, |
1803 | &ordered_extent->list); | |
1804 | ||
1ef30be1 | 1805 | ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent); |
a39f7521 | 1806 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) { |
2115133f | 1807 | ret = btrfs_update_inode_fallback(trans, root, inode); |
1ef30be1 JB |
1808 | BUG_ON(ret); |
1809 | } | |
1810 | ret = 0; | |
c2167754 | 1811 | out: |
5b0e95bf | 1812 | if (root != root->fs_info->tree_root) |
0cb59c99 | 1813 | btrfs_delalloc_release_metadata(inode, ordered_extent->len); |
5b0e95bf JB |
1814 | if (trans) { |
1815 | if (nolock) | |
0cb59c99 | 1816 | btrfs_end_transaction_nolock(trans, root); |
5b0e95bf | 1817 | else |
0cb59c99 JB |
1818 | btrfs_end_transaction(trans, root); |
1819 | } | |
1820 | ||
e6dcd2dc CM |
1821 | /* once for us */ |
1822 | btrfs_put_ordered_extent(ordered_extent); | |
1823 | /* once for the tree */ | |
1824 | btrfs_put_ordered_extent(ordered_extent); | |
1825 | ||
e6dcd2dc CM |
1826 | return 0; |
1827 | } | |
1828 | ||
b2950863 | 1829 | static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end, |
211f90e6 CM |
1830 | struct extent_state *state, int uptodate) |
1831 | { | |
1abe9b8a | 1832 | trace_btrfs_writepage_end_io_hook(page, start, end, uptodate); |
1833 | ||
8b62b72b | 1834 | ClearPagePrivate2(page); |
211f90e6 CM |
1835 | return btrfs_finish_ordered_io(page->mapping->host, start, end); |
1836 | } | |
1837 | ||
d352ac68 CM |
1838 | /* |
1839 | * when reads are done, we need to check csums to verify the data is correct | |
4a54c8c1 JS |
1840 | * if there's a match, we allow the bio to finish. If not, the code in |
1841 | * extent_io.c will try to find good copies for us. | |
d352ac68 | 1842 | */ |
b2950863 | 1843 | static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end, |
70dec807 | 1844 | struct extent_state *state) |
07157aac | 1845 | { |
35ebb934 | 1846 | size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT); |
07157aac | 1847 | struct inode *inode = page->mapping->host; |
d1310b2e | 1848 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
07157aac | 1849 | char *kaddr; |
aadfeb6e | 1850 | u64 private = ~(u32)0; |
07157aac | 1851 | int ret; |
ff79f819 CM |
1852 | struct btrfs_root *root = BTRFS_I(inode)->root; |
1853 | u32 csum = ~(u32)0; | |
d1310b2e | 1854 | |
d20f7043 CM |
1855 | if (PageChecked(page)) { |
1856 | ClearPageChecked(page); | |
1857 | goto good; | |
1858 | } | |
6cbff00f CH |
1859 | |
1860 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) | |
08d2f347 | 1861 | goto good; |
17d217fe YZ |
1862 | |
1863 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID && | |
9655d298 | 1864 | test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) { |
17d217fe YZ |
1865 | clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM, |
1866 | GFP_NOFS); | |
b6cda9bc | 1867 | return 0; |
17d217fe | 1868 | } |
d20f7043 | 1869 | |
c2e639f0 | 1870 | if (state && state->start == start) { |
70dec807 CM |
1871 | private = state->private; |
1872 | ret = 0; | |
1873 | } else { | |
1874 | ret = get_state_private(io_tree, start, &private); | |
1875 | } | |
9ab86c8e | 1876 | kaddr = kmap_atomic(page, KM_USER0); |
d397712b | 1877 | if (ret) |
07157aac | 1878 | goto zeroit; |
d397712b | 1879 | |
ff79f819 CM |
1880 | csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1); |
1881 | btrfs_csum_final(csum, (char *)&csum); | |
d397712b | 1882 | if (csum != private) |
07157aac | 1883 | goto zeroit; |
d397712b | 1884 | |
9ab86c8e | 1885 | kunmap_atomic(kaddr, KM_USER0); |
d20f7043 | 1886 | good: |
07157aac CM |
1887 | return 0; |
1888 | ||
1889 | zeroit: | |
945d8962 | 1890 | printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u " |
33345d01 LZ |
1891 | "private %llu\n", |
1892 | (unsigned long long)btrfs_ino(page->mapping->host), | |
193f284d CM |
1893 | (unsigned long long)start, csum, |
1894 | (unsigned long long)private); | |
db94535d CM |
1895 | memset(kaddr + offset, 1, end - start + 1); |
1896 | flush_dcache_page(page); | |
9ab86c8e | 1897 | kunmap_atomic(kaddr, KM_USER0); |
3b951516 CM |
1898 | if (private == 0) |
1899 | return 0; | |
7e38326f | 1900 | return -EIO; |
07157aac | 1901 | } |
b888db2b | 1902 | |
24bbcf04 YZ |
1903 | struct delayed_iput { |
1904 | struct list_head list; | |
1905 | struct inode *inode; | |
1906 | }; | |
1907 | ||
1908 | void btrfs_add_delayed_iput(struct inode *inode) | |
1909 | { | |
1910 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
1911 | struct delayed_iput *delayed; | |
1912 | ||
1913 | if (atomic_add_unless(&inode->i_count, -1, 1)) | |
1914 | return; | |
1915 | ||
1916 | delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL); | |
1917 | delayed->inode = inode; | |
1918 | ||
1919 | spin_lock(&fs_info->delayed_iput_lock); | |
1920 | list_add_tail(&delayed->list, &fs_info->delayed_iputs); | |
1921 | spin_unlock(&fs_info->delayed_iput_lock); | |
1922 | } | |
1923 | ||
1924 | void btrfs_run_delayed_iputs(struct btrfs_root *root) | |
1925 | { | |
1926 | LIST_HEAD(list); | |
1927 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1928 | struct delayed_iput *delayed; | |
1929 | int empty; | |
1930 | ||
1931 | spin_lock(&fs_info->delayed_iput_lock); | |
1932 | empty = list_empty(&fs_info->delayed_iputs); | |
1933 | spin_unlock(&fs_info->delayed_iput_lock); | |
1934 | if (empty) | |
1935 | return; | |
1936 | ||
1937 | down_read(&root->fs_info->cleanup_work_sem); | |
1938 | spin_lock(&fs_info->delayed_iput_lock); | |
1939 | list_splice_init(&fs_info->delayed_iputs, &list); | |
1940 | spin_unlock(&fs_info->delayed_iput_lock); | |
1941 | ||
1942 | while (!list_empty(&list)) { | |
1943 | delayed = list_entry(list.next, struct delayed_iput, list); | |
1944 | list_del(&delayed->list); | |
1945 | iput(delayed->inode); | |
1946 | kfree(delayed); | |
1947 | } | |
1948 | up_read(&root->fs_info->cleanup_work_sem); | |
1949 | } | |
1950 | ||
d68fc57b YZ |
1951 | enum btrfs_orphan_cleanup_state { |
1952 | ORPHAN_CLEANUP_STARTED = 1, | |
1953 | ORPHAN_CLEANUP_DONE = 2, | |
1954 | }; | |
1955 | ||
1956 | /* | |
1957 | * This is called in transaction commmit time. If there are no orphan | |
1958 | * files in the subvolume, it removes orphan item and frees block_rsv | |
1959 | * structure. | |
1960 | */ | |
1961 | void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans, | |
1962 | struct btrfs_root *root) | |
1963 | { | |
90290e19 | 1964 | struct btrfs_block_rsv *block_rsv; |
d68fc57b YZ |
1965 | int ret; |
1966 | ||
1967 | if (!list_empty(&root->orphan_list) || | |
1968 | root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) | |
1969 | return; | |
1970 | ||
90290e19 JB |
1971 | spin_lock(&root->orphan_lock); |
1972 | if (!list_empty(&root->orphan_list)) { | |
1973 | spin_unlock(&root->orphan_lock); | |
1974 | return; | |
1975 | } | |
1976 | ||
1977 | if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) { | |
1978 | spin_unlock(&root->orphan_lock); | |
1979 | return; | |
1980 | } | |
1981 | ||
1982 | block_rsv = root->orphan_block_rsv; | |
1983 | root->orphan_block_rsv = NULL; | |
1984 | spin_unlock(&root->orphan_lock); | |
1985 | ||
d68fc57b YZ |
1986 | if (root->orphan_item_inserted && |
1987 | btrfs_root_refs(&root->root_item) > 0) { | |
1988 | ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root, | |
1989 | root->root_key.objectid); | |
1990 | BUG_ON(ret); | |
1991 | root->orphan_item_inserted = 0; | |
1992 | } | |
1993 | ||
90290e19 JB |
1994 | if (block_rsv) { |
1995 | WARN_ON(block_rsv->size > 0); | |
1996 | btrfs_free_block_rsv(root, block_rsv); | |
d68fc57b YZ |
1997 | } |
1998 | } | |
1999 | ||
7b128766 JB |
2000 | /* |
2001 | * This creates an orphan entry for the given inode in case something goes | |
2002 | * wrong in the middle of an unlink/truncate. | |
d68fc57b YZ |
2003 | * |
2004 | * NOTE: caller of this function should reserve 5 units of metadata for | |
2005 | * this function. | |
7b128766 JB |
2006 | */ |
2007 | int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode) | |
2008 | { | |
2009 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2010 | struct btrfs_block_rsv *block_rsv = NULL; |
2011 | int reserve = 0; | |
2012 | int insert = 0; | |
2013 | int ret; | |
7b128766 | 2014 | |
d68fc57b YZ |
2015 | if (!root->orphan_block_rsv) { |
2016 | block_rsv = btrfs_alloc_block_rsv(root); | |
b532402e TI |
2017 | if (!block_rsv) |
2018 | return -ENOMEM; | |
d68fc57b | 2019 | } |
7b128766 | 2020 | |
d68fc57b YZ |
2021 | spin_lock(&root->orphan_lock); |
2022 | if (!root->orphan_block_rsv) { | |
2023 | root->orphan_block_rsv = block_rsv; | |
2024 | } else if (block_rsv) { | |
2025 | btrfs_free_block_rsv(root, block_rsv); | |
2026 | block_rsv = NULL; | |
7b128766 | 2027 | } |
7b128766 | 2028 | |
d68fc57b YZ |
2029 | if (list_empty(&BTRFS_I(inode)->i_orphan)) { |
2030 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2031 | #if 0 | |
2032 | /* | |
2033 | * For proper ENOSPC handling, we should do orphan | |
2034 | * cleanup when mounting. But this introduces backward | |
2035 | * compatibility issue. | |
2036 | */ | |
2037 | if (!xchg(&root->orphan_item_inserted, 1)) | |
2038 | insert = 2; | |
2039 | else | |
2040 | insert = 1; | |
2041 | #endif | |
2042 | insert = 1; | |
7b128766 JB |
2043 | } |
2044 | ||
d68fc57b YZ |
2045 | if (!BTRFS_I(inode)->orphan_meta_reserved) { |
2046 | BTRFS_I(inode)->orphan_meta_reserved = 1; | |
2047 | reserve = 1; | |
2048 | } | |
2049 | spin_unlock(&root->orphan_lock); | |
7b128766 | 2050 | |
d68fc57b YZ |
2051 | /* grab metadata reservation from transaction handle */ |
2052 | if (reserve) { | |
2053 | ret = btrfs_orphan_reserve_metadata(trans, inode); | |
2054 | BUG_ON(ret); | |
2055 | } | |
7b128766 | 2056 | |
d68fc57b YZ |
2057 | /* insert an orphan item to track this unlinked/truncated file */ |
2058 | if (insert >= 1) { | |
33345d01 | 2059 | ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode)); |
ee4d89f0 | 2060 | BUG_ON(ret && ret != -EEXIST); |
d68fc57b YZ |
2061 | } |
2062 | ||
2063 | /* insert an orphan item to track subvolume contains orphan files */ | |
2064 | if (insert >= 2) { | |
2065 | ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root, | |
2066 | root->root_key.objectid); | |
2067 | BUG_ON(ret); | |
2068 | } | |
2069 | return 0; | |
7b128766 JB |
2070 | } |
2071 | ||
2072 | /* | |
2073 | * We have done the truncate/delete so we can go ahead and remove the orphan | |
2074 | * item for this particular inode. | |
2075 | */ | |
2076 | int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode) | |
2077 | { | |
2078 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
d68fc57b YZ |
2079 | int delete_item = 0; |
2080 | int release_rsv = 0; | |
7b128766 JB |
2081 | int ret = 0; |
2082 | ||
d68fc57b YZ |
2083 | spin_lock(&root->orphan_lock); |
2084 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { | |
2085 | list_del_init(&BTRFS_I(inode)->i_orphan); | |
2086 | delete_item = 1; | |
7b128766 JB |
2087 | } |
2088 | ||
d68fc57b YZ |
2089 | if (BTRFS_I(inode)->orphan_meta_reserved) { |
2090 | BTRFS_I(inode)->orphan_meta_reserved = 0; | |
2091 | release_rsv = 1; | |
7b128766 | 2092 | } |
d68fc57b | 2093 | spin_unlock(&root->orphan_lock); |
7b128766 | 2094 | |
d68fc57b | 2095 | if (trans && delete_item) { |
33345d01 | 2096 | ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode)); |
d68fc57b YZ |
2097 | BUG_ON(ret); |
2098 | } | |
7b128766 | 2099 | |
d68fc57b YZ |
2100 | if (release_rsv) |
2101 | btrfs_orphan_release_metadata(inode); | |
7b128766 | 2102 | |
d68fc57b | 2103 | return 0; |
7b128766 JB |
2104 | } |
2105 | ||
2106 | /* | |
2107 | * this cleans up any orphans that may be left on the list from the last use | |
2108 | * of this root. | |
2109 | */ | |
66b4ffd1 | 2110 | int btrfs_orphan_cleanup(struct btrfs_root *root) |
7b128766 JB |
2111 | { |
2112 | struct btrfs_path *path; | |
2113 | struct extent_buffer *leaf; | |
7b128766 JB |
2114 | struct btrfs_key key, found_key; |
2115 | struct btrfs_trans_handle *trans; | |
2116 | struct inode *inode; | |
8f6d7f4f | 2117 | u64 last_objectid = 0; |
7b128766 JB |
2118 | int ret = 0, nr_unlink = 0, nr_truncate = 0; |
2119 | ||
d68fc57b | 2120 | if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED)) |
66b4ffd1 | 2121 | return 0; |
c71bf099 YZ |
2122 | |
2123 | path = btrfs_alloc_path(); | |
66b4ffd1 JB |
2124 | if (!path) { |
2125 | ret = -ENOMEM; | |
2126 | goto out; | |
2127 | } | |
7b128766 JB |
2128 | path->reada = -1; |
2129 | ||
2130 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
2131 | btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY); | |
2132 | key.offset = (u64)-1; | |
2133 | ||
7b128766 JB |
2134 | while (1) { |
2135 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
66b4ffd1 JB |
2136 | if (ret < 0) |
2137 | goto out; | |
7b128766 JB |
2138 | |
2139 | /* | |
2140 | * if ret == 0 means we found what we were searching for, which | |
25985edc | 2141 | * is weird, but possible, so only screw with path if we didn't |
7b128766 JB |
2142 | * find the key and see if we have stuff that matches |
2143 | */ | |
2144 | if (ret > 0) { | |
66b4ffd1 | 2145 | ret = 0; |
7b128766 JB |
2146 | if (path->slots[0] == 0) |
2147 | break; | |
2148 | path->slots[0]--; | |
2149 | } | |
2150 | ||
2151 | /* pull out the item */ | |
2152 | leaf = path->nodes[0]; | |
7b128766 JB |
2153 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
2154 | ||
2155 | /* make sure the item matches what we want */ | |
2156 | if (found_key.objectid != BTRFS_ORPHAN_OBJECTID) | |
2157 | break; | |
2158 | if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY) | |
2159 | break; | |
2160 | ||
2161 | /* release the path since we're done with it */ | |
b3b4aa74 | 2162 | btrfs_release_path(path); |
7b128766 JB |
2163 | |
2164 | /* | |
2165 | * this is where we are basically btrfs_lookup, without the | |
2166 | * crossing root thing. we store the inode number in the | |
2167 | * offset of the orphan item. | |
2168 | */ | |
8f6d7f4f JB |
2169 | |
2170 | if (found_key.offset == last_objectid) { | |
2171 | printk(KERN_ERR "btrfs: Error removing orphan entry, " | |
2172 | "stopping orphan cleanup\n"); | |
2173 | ret = -EINVAL; | |
2174 | goto out; | |
2175 | } | |
2176 | ||
2177 | last_objectid = found_key.offset; | |
2178 | ||
5d4f98a2 YZ |
2179 | found_key.objectid = found_key.offset; |
2180 | found_key.type = BTRFS_INODE_ITEM_KEY; | |
2181 | found_key.offset = 0; | |
73f73415 | 2182 | inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL); |
a8c9e576 JB |
2183 | ret = PTR_RET(inode); |
2184 | if (ret && ret != -ESTALE) | |
66b4ffd1 | 2185 | goto out; |
7b128766 | 2186 | |
f8e9e0b0 AJ |
2187 | if (ret == -ESTALE && root == root->fs_info->tree_root) { |
2188 | struct btrfs_root *dead_root; | |
2189 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2190 | int is_dead_root = 0; | |
2191 | ||
2192 | /* | |
2193 | * this is an orphan in the tree root. Currently these | |
2194 | * could come from 2 sources: | |
2195 | * a) a snapshot deletion in progress | |
2196 | * b) a free space cache inode | |
2197 | * We need to distinguish those two, as the snapshot | |
2198 | * orphan must not get deleted. | |
2199 | * find_dead_roots already ran before us, so if this | |
2200 | * is a snapshot deletion, we should find the root | |
2201 | * in the dead_roots list | |
2202 | */ | |
2203 | spin_lock(&fs_info->trans_lock); | |
2204 | list_for_each_entry(dead_root, &fs_info->dead_roots, | |
2205 | root_list) { | |
2206 | if (dead_root->root_key.objectid == | |
2207 | found_key.objectid) { | |
2208 | is_dead_root = 1; | |
2209 | break; | |
2210 | } | |
2211 | } | |
2212 | spin_unlock(&fs_info->trans_lock); | |
2213 | if (is_dead_root) { | |
2214 | /* prevent this orphan from being found again */ | |
2215 | key.offset = found_key.objectid - 1; | |
2216 | continue; | |
2217 | } | |
2218 | } | |
7b128766 | 2219 | /* |
a8c9e576 JB |
2220 | * Inode is already gone but the orphan item is still there, |
2221 | * kill the orphan item. | |
7b128766 | 2222 | */ |
a8c9e576 JB |
2223 | if (ret == -ESTALE) { |
2224 | trans = btrfs_start_transaction(root, 1); | |
66b4ffd1 JB |
2225 | if (IS_ERR(trans)) { |
2226 | ret = PTR_ERR(trans); | |
2227 | goto out; | |
2228 | } | |
a8c9e576 JB |
2229 | ret = btrfs_del_orphan_item(trans, root, |
2230 | found_key.objectid); | |
2231 | BUG_ON(ret); | |
5b21f2ed | 2232 | btrfs_end_transaction(trans, root); |
7b128766 JB |
2233 | continue; |
2234 | } | |
2235 | ||
a8c9e576 JB |
2236 | /* |
2237 | * add this inode to the orphan list so btrfs_orphan_del does | |
2238 | * the proper thing when we hit it | |
2239 | */ | |
2240 | spin_lock(&root->orphan_lock); | |
2241 | list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list); | |
2242 | spin_unlock(&root->orphan_lock); | |
2243 | ||
7b128766 JB |
2244 | /* if we have links, this was a truncate, lets do that */ |
2245 | if (inode->i_nlink) { | |
a41ad394 JB |
2246 | if (!S_ISREG(inode->i_mode)) { |
2247 | WARN_ON(1); | |
2248 | iput(inode); | |
2249 | continue; | |
2250 | } | |
7b128766 | 2251 | nr_truncate++; |
66b4ffd1 | 2252 | ret = btrfs_truncate(inode); |
7b128766 JB |
2253 | } else { |
2254 | nr_unlink++; | |
2255 | } | |
2256 | ||
2257 | /* this will do delete_inode and everything for us */ | |
2258 | iput(inode); | |
66b4ffd1 JB |
2259 | if (ret) |
2260 | goto out; | |
7b128766 | 2261 | } |
3254c876 MX |
2262 | /* release the path since we're done with it */ |
2263 | btrfs_release_path(path); | |
2264 | ||
d68fc57b YZ |
2265 | root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE; |
2266 | ||
2267 | if (root->orphan_block_rsv) | |
2268 | btrfs_block_rsv_release(root, root->orphan_block_rsv, | |
2269 | (u64)-1); | |
2270 | ||
2271 | if (root->orphan_block_rsv || root->orphan_item_inserted) { | |
7a7eaa40 | 2272 | trans = btrfs_join_transaction(root); |
66b4ffd1 JB |
2273 | if (!IS_ERR(trans)) |
2274 | btrfs_end_transaction(trans, root); | |
d68fc57b | 2275 | } |
7b128766 JB |
2276 | |
2277 | if (nr_unlink) | |
2278 | printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink); | |
2279 | if (nr_truncate) | |
2280 | printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate); | |
66b4ffd1 JB |
2281 | |
2282 | out: | |
2283 | if (ret) | |
2284 | printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret); | |
2285 | btrfs_free_path(path); | |
2286 | return ret; | |
7b128766 JB |
2287 | } |
2288 | ||
46a53cca CM |
2289 | /* |
2290 | * very simple check to peek ahead in the leaf looking for xattrs. If we | |
2291 | * don't find any xattrs, we know there can't be any acls. | |
2292 | * | |
2293 | * slot is the slot the inode is in, objectid is the objectid of the inode | |
2294 | */ | |
2295 | static noinline int acls_after_inode_item(struct extent_buffer *leaf, | |
2296 | int slot, u64 objectid) | |
2297 | { | |
2298 | u32 nritems = btrfs_header_nritems(leaf); | |
2299 | struct btrfs_key found_key; | |
2300 | int scanned = 0; | |
2301 | ||
2302 | slot++; | |
2303 | while (slot < nritems) { | |
2304 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2305 | ||
2306 | /* we found a different objectid, there must not be acls */ | |
2307 | if (found_key.objectid != objectid) | |
2308 | return 0; | |
2309 | ||
2310 | /* we found an xattr, assume we've got an acl */ | |
2311 | if (found_key.type == BTRFS_XATTR_ITEM_KEY) | |
2312 | return 1; | |
2313 | ||
2314 | /* | |
2315 | * we found a key greater than an xattr key, there can't | |
2316 | * be any acls later on | |
2317 | */ | |
2318 | if (found_key.type > BTRFS_XATTR_ITEM_KEY) | |
2319 | return 0; | |
2320 | ||
2321 | slot++; | |
2322 | scanned++; | |
2323 | ||
2324 | /* | |
2325 | * it goes inode, inode backrefs, xattrs, extents, | |
2326 | * so if there are a ton of hard links to an inode there can | |
2327 | * be a lot of backrefs. Don't waste time searching too hard, | |
2328 | * this is just an optimization | |
2329 | */ | |
2330 | if (scanned >= 8) | |
2331 | break; | |
2332 | } | |
2333 | /* we hit the end of the leaf before we found an xattr or | |
2334 | * something larger than an xattr. We have to assume the inode | |
2335 | * has acls | |
2336 | */ | |
2337 | return 1; | |
2338 | } | |
2339 | ||
d352ac68 CM |
2340 | /* |
2341 | * read an inode from the btree into the in-memory inode | |
2342 | */ | |
5d4f98a2 | 2343 | static void btrfs_read_locked_inode(struct inode *inode) |
39279cc3 CM |
2344 | { |
2345 | struct btrfs_path *path; | |
5f39d397 | 2346 | struct extent_buffer *leaf; |
39279cc3 | 2347 | struct btrfs_inode_item *inode_item; |
0b86a832 | 2348 | struct btrfs_timespec *tspec; |
39279cc3 CM |
2349 | struct btrfs_root *root = BTRFS_I(inode)->root; |
2350 | struct btrfs_key location; | |
46a53cca | 2351 | int maybe_acls; |
618e21d5 | 2352 | u32 rdev; |
39279cc3 | 2353 | int ret; |
2f7e33d4 MX |
2354 | bool filled = false; |
2355 | ||
2356 | ret = btrfs_fill_inode(inode, &rdev); | |
2357 | if (!ret) | |
2358 | filled = true; | |
39279cc3 CM |
2359 | |
2360 | path = btrfs_alloc_path(); | |
1748f843 MF |
2361 | if (!path) |
2362 | goto make_bad; | |
2363 | ||
d90c7321 | 2364 | path->leave_spinning = 1; |
39279cc3 | 2365 | memcpy(&location, &BTRFS_I(inode)->location, sizeof(location)); |
dc17ff8f | 2366 | |
39279cc3 | 2367 | ret = btrfs_lookup_inode(NULL, root, path, &location, 0); |
5f39d397 | 2368 | if (ret) |
39279cc3 | 2369 | goto make_bad; |
39279cc3 | 2370 | |
5f39d397 | 2371 | leaf = path->nodes[0]; |
2f7e33d4 MX |
2372 | |
2373 | if (filled) | |
2374 | goto cache_acl; | |
2375 | ||
5f39d397 CM |
2376 | inode_item = btrfs_item_ptr(leaf, path->slots[0], |
2377 | struct btrfs_inode_item); | |
5f39d397 | 2378 | inode->i_mode = btrfs_inode_mode(leaf, inode_item); |
bfe86848 | 2379 | set_nlink(inode, btrfs_inode_nlink(leaf, inode_item)); |
5f39d397 CM |
2380 | inode->i_uid = btrfs_inode_uid(leaf, inode_item); |
2381 | inode->i_gid = btrfs_inode_gid(leaf, inode_item); | |
dbe674a9 | 2382 | btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item)); |
5f39d397 CM |
2383 | |
2384 | tspec = btrfs_inode_atime(inode_item); | |
2385 | inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2386 | inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2387 | ||
2388 | tspec = btrfs_inode_mtime(inode_item); | |
2389 | inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2390 | inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2391 | ||
2392 | tspec = btrfs_inode_ctime(inode_item); | |
2393 | inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec); | |
2394 | inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); | |
2395 | ||
a76a3cd4 | 2396 | inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item)); |
e02119d5 | 2397 | BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item); |
c3027eb5 | 2398 | BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item); |
e02119d5 | 2399 | inode->i_generation = BTRFS_I(inode)->generation; |
618e21d5 | 2400 | inode->i_rdev = 0; |
5f39d397 CM |
2401 | rdev = btrfs_inode_rdev(leaf, inode_item); |
2402 | ||
aec7477b | 2403 | BTRFS_I(inode)->index_cnt = (u64)-1; |
d2fb3437 | 2404 | BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item); |
2f7e33d4 | 2405 | cache_acl: |
46a53cca CM |
2406 | /* |
2407 | * try to precache a NULL acl entry for files that don't have | |
2408 | * any xattrs or acls | |
2409 | */ | |
33345d01 LZ |
2410 | maybe_acls = acls_after_inode_item(leaf, path->slots[0], |
2411 | btrfs_ino(inode)); | |
72c04902 AV |
2412 | if (!maybe_acls) |
2413 | cache_no_acl(inode); | |
46a53cca | 2414 | |
39279cc3 | 2415 | btrfs_free_path(path); |
39279cc3 | 2416 | |
39279cc3 | 2417 | switch (inode->i_mode & S_IFMT) { |
39279cc3 CM |
2418 | case S_IFREG: |
2419 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 2420 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 2421 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 CM |
2422 | inode->i_fop = &btrfs_file_operations; |
2423 | inode->i_op = &btrfs_file_inode_operations; | |
2424 | break; | |
2425 | case S_IFDIR: | |
2426 | inode->i_fop = &btrfs_dir_file_operations; | |
2427 | if (root == root->fs_info->tree_root) | |
2428 | inode->i_op = &btrfs_dir_ro_inode_operations; | |
2429 | else | |
2430 | inode->i_op = &btrfs_dir_inode_operations; | |
2431 | break; | |
2432 | case S_IFLNK: | |
2433 | inode->i_op = &btrfs_symlink_inode_operations; | |
2434 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 2435 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
39279cc3 | 2436 | break; |
618e21d5 | 2437 | default: |
0279b4cd | 2438 | inode->i_op = &btrfs_special_inode_operations; |
618e21d5 JB |
2439 | init_special_inode(inode, inode->i_mode, rdev); |
2440 | break; | |
39279cc3 | 2441 | } |
6cbff00f CH |
2442 | |
2443 | btrfs_update_iflags(inode); | |
39279cc3 CM |
2444 | return; |
2445 | ||
2446 | make_bad: | |
39279cc3 | 2447 | btrfs_free_path(path); |
39279cc3 CM |
2448 | make_bad_inode(inode); |
2449 | } | |
2450 | ||
d352ac68 CM |
2451 | /* |
2452 | * given a leaf and an inode, copy the inode fields into the leaf | |
2453 | */ | |
e02119d5 CM |
2454 | static void fill_inode_item(struct btrfs_trans_handle *trans, |
2455 | struct extent_buffer *leaf, | |
5f39d397 | 2456 | struct btrfs_inode_item *item, |
39279cc3 CM |
2457 | struct inode *inode) |
2458 | { | |
5f39d397 CM |
2459 | btrfs_set_inode_uid(leaf, item, inode->i_uid); |
2460 | btrfs_set_inode_gid(leaf, item, inode->i_gid); | |
dbe674a9 | 2461 | btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size); |
5f39d397 CM |
2462 | btrfs_set_inode_mode(leaf, item, inode->i_mode); |
2463 | btrfs_set_inode_nlink(leaf, item, inode->i_nlink); | |
2464 | ||
2465 | btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item), | |
2466 | inode->i_atime.tv_sec); | |
2467 | btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item), | |
2468 | inode->i_atime.tv_nsec); | |
2469 | ||
2470 | btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item), | |
2471 | inode->i_mtime.tv_sec); | |
2472 | btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item), | |
2473 | inode->i_mtime.tv_nsec); | |
2474 | ||
2475 | btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item), | |
2476 | inode->i_ctime.tv_sec); | |
2477 | btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item), | |
2478 | inode->i_ctime.tv_nsec); | |
2479 | ||
a76a3cd4 | 2480 | btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode)); |
e02119d5 | 2481 | btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation); |
c3027eb5 | 2482 | btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence); |
e02119d5 | 2483 | btrfs_set_inode_transid(leaf, item, trans->transid); |
5f39d397 | 2484 | btrfs_set_inode_rdev(leaf, item, inode->i_rdev); |
b98b6767 | 2485 | btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); |
d82a6f1d | 2486 | btrfs_set_inode_block_group(leaf, item, 0); |
39279cc3 CM |
2487 | } |
2488 | ||
d352ac68 CM |
2489 | /* |
2490 | * copy everything in the in-memory inode into the btree. | |
2491 | */ | |
2115133f | 2492 | static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans, |
d397712b | 2493 | struct btrfs_root *root, struct inode *inode) |
39279cc3 CM |
2494 | { |
2495 | struct btrfs_inode_item *inode_item; | |
2496 | struct btrfs_path *path; | |
5f39d397 | 2497 | struct extent_buffer *leaf; |
39279cc3 CM |
2498 | int ret; |
2499 | ||
2500 | path = btrfs_alloc_path(); | |
16cdcec7 MX |
2501 | if (!path) |
2502 | return -ENOMEM; | |
2503 | ||
b9473439 | 2504 | path->leave_spinning = 1; |
16cdcec7 MX |
2505 | ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location, |
2506 | 1); | |
39279cc3 CM |
2507 | if (ret) { |
2508 | if (ret > 0) | |
2509 | ret = -ENOENT; | |
2510 | goto failed; | |
2511 | } | |
2512 | ||
b4ce94de | 2513 | btrfs_unlock_up_safe(path, 1); |
5f39d397 CM |
2514 | leaf = path->nodes[0]; |
2515 | inode_item = btrfs_item_ptr(leaf, path->slots[0], | |
16cdcec7 | 2516 | struct btrfs_inode_item); |
39279cc3 | 2517 | |
e02119d5 | 2518 | fill_inode_item(trans, leaf, inode_item, inode); |
5f39d397 | 2519 | btrfs_mark_buffer_dirty(leaf); |
15ee9bc7 | 2520 | btrfs_set_inode_last_trans(trans, inode); |
39279cc3 CM |
2521 | ret = 0; |
2522 | failed: | |
39279cc3 CM |
2523 | btrfs_free_path(path); |
2524 | return ret; | |
2525 | } | |
2526 | ||
2115133f CM |
2527 | /* |
2528 | * copy everything in the in-memory inode into the btree. | |
2529 | */ | |
2530 | noinline int btrfs_update_inode(struct btrfs_trans_handle *trans, | |
2531 | struct btrfs_root *root, struct inode *inode) | |
2532 | { | |
2533 | int ret; | |
2534 | ||
2535 | /* | |
2536 | * If the inode is a free space inode, we can deadlock during commit | |
2537 | * if we put it into the delayed code. | |
2538 | * | |
2539 | * The data relocation inode should also be directly updated | |
2540 | * without delay | |
2541 | */ | |
2542 | if (!btrfs_is_free_space_inode(root, inode) | |
2543 | && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
2544 | ret = btrfs_delayed_update_inode(trans, root, inode); | |
2545 | if (!ret) | |
2546 | btrfs_set_inode_last_trans(trans, inode); | |
2547 | return ret; | |
2548 | } | |
2549 | ||
2550 | return btrfs_update_inode_item(trans, root, inode); | |
2551 | } | |
2552 | ||
2553 | static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans, | |
2554 | struct btrfs_root *root, struct inode *inode) | |
2555 | { | |
2556 | int ret; | |
2557 | ||
2558 | ret = btrfs_update_inode(trans, root, inode); | |
2559 | if (ret == -ENOSPC) | |
2560 | return btrfs_update_inode_item(trans, root, inode); | |
2561 | return ret; | |
2562 | } | |
2563 | ||
d352ac68 CM |
2564 | /* |
2565 | * unlink helper that gets used here in inode.c and in the tree logging | |
2566 | * recovery code. It remove a link in a directory with a given name, and | |
2567 | * also drops the back refs in the inode to the directory | |
2568 | */ | |
92986796 AV |
2569 | static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2570 | struct btrfs_root *root, | |
2571 | struct inode *dir, struct inode *inode, | |
2572 | const char *name, int name_len) | |
39279cc3 CM |
2573 | { |
2574 | struct btrfs_path *path; | |
39279cc3 | 2575 | int ret = 0; |
5f39d397 | 2576 | struct extent_buffer *leaf; |
39279cc3 | 2577 | struct btrfs_dir_item *di; |
5f39d397 | 2578 | struct btrfs_key key; |
aec7477b | 2579 | u64 index; |
33345d01 LZ |
2580 | u64 ino = btrfs_ino(inode); |
2581 | u64 dir_ino = btrfs_ino(dir); | |
39279cc3 CM |
2582 | |
2583 | path = btrfs_alloc_path(); | |
54aa1f4d CM |
2584 | if (!path) { |
2585 | ret = -ENOMEM; | |
554233a6 | 2586 | goto out; |
54aa1f4d CM |
2587 | } |
2588 | ||
b9473439 | 2589 | path->leave_spinning = 1; |
33345d01 | 2590 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
39279cc3 CM |
2591 | name, name_len, -1); |
2592 | if (IS_ERR(di)) { | |
2593 | ret = PTR_ERR(di); | |
2594 | goto err; | |
2595 | } | |
2596 | if (!di) { | |
2597 | ret = -ENOENT; | |
2598 | goto err; | |
2599 | } | |
5f39d397 CM |
2600 | leaf = path->nodes[0]; |
2601 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
39279cc3 | 2602 | ret = btrfs_delete_one_dir_name(trans, root, path, di); |
54aa1f4d CM |
2603 | if (ret) |
2604 | goto err; | |
b3b4aa74 | 2605 | btrfs_release_path(path); |
39279cc3 | 2606 | |
33345d01 LZ |
2607 | ret = btrfs_del_inode_ref(trans, root, name, name_len, ino, |
2608 | dir_ino, &index); | |
aec7477b | 2609 | if (ret) { |
d397712b | 2610 | printk(KERN_INFO "btrfs failed to delete reference to %.*s, " |
33345d01 LZ |
2611 | "inode %llu parent %llu\n", name_len, name, |
2612 | (unsigned long long)ino, (unsigned long long)dir_ino); | |
aec7477b JB |
2613 | goto err; |
2614 | } | |
2615 | ||
16cdcec7 MX |
2616 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
2617 | if (ret) | |
39279cc3 | 2618 | goto err; |
39279cc3 | 2619 | |
e02119d5 | 2620 | ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, |
33345d01 | 2621 | inode, dir_ino); |
49eb7e46 | 2622 | BUG_ON(ret != 0 && ret != -ENOENT); |
e02119d5 CM |
2623 | |
2624 | ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, | |
2625 | dir, index); | |
6418c961 CM |
2626 | if (ret == -ENOENT) |
2627 | ret = 0; | |
39279cc3 CM |
2628 | err: |
2629 | btrfs_free_path(path); | |
e02119d5 CM |
2630 | if (ret) |
2631 | goto out; | |
2632 | ||
2633 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2634 | inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2635 | btrfs_update_inode(trans, root, dir); | |
e02119d5 | 2636 | out: |
39279cc3 CM |
2637 | return ret; |
2638 | } | |
2639 | ||
92986796 AV |
2640 | int btrfs_unlink_inode(struct btrfs_trans_handle *trans, |
2641 | struct btrfs_root *root, | |
2642 | struct inode *dir, struct inode *inode, | |
2643 | const char *name, int name_len) | |
2644 | { | |
2645 | int ret; | |
2646 | ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len); | |
2647 | if (!ret) { | |
2648 | btrfs_drop_nlink(inode); | |
2649 | ret = btrfs_update_inode(trans, root, inode); | |
2650 | } | |
2651 | return ret; | |
2652 | } | |
2653 | ||
2654 | ||
a22285a6 YZ |
2655 | /* helper to check if there is any shared block in the path */ |
2656 | static int check_path_shared(struct btrfs_root *root, | |
2657 | struct btrfs_path *path) | |
39279cc3 | 2658 | { |
a22285a6 YZ |
2659 | struct extent_buffer *eb; |
2660 | int level; | |
0e4dcbef | 2661 | u64 refs = 1; |
5df6a9f6 | 2662 | |
a22285a6 | 2663 | for (level = 0; level < BTRFS_MAX_LEVEL; level++) { |
dedefd72 JB |
2664 | int ret; |
2665 | ||
a22285a6 YZ |
2666 | if (!path->nodes[level]) |
2667 | break; | |
2668 | eb = path->nodes[level]; | |
2669 | if (!btrfs_block_can_be_shared(root, eb)) | |
2670 | continue; | |
2671 | ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len, | |
2672 | &refs, NULL); | |
2673 | if (refs > 1) | |
2674 | return 1; | |
5df6a9f6 | 2675 | } |
dedefd72 | 2676 | return 0; |
39279cc3 CM |
2677 | } |
2678 | ||
a22285a6 YZ |
2679 | /* |
2680 | * helper to start transaction for unlink and rmdir. | |
2681 | * | |
2682 | * unlink and rmdir are special in btrfs, they do not always free space. | |
2683 | * so in enospc case, we should make sure they will free space before | |
2684 | * allowing them to use the global metadata reservation. | |
2685 | */ | |
2686 | static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir, | |
2687 | struct dentry *dentry) | |
4df27c4d | 2688 | { |
39279cc3 | 2689 | struct btrfs_trans_handle *trans; |
a22285a6 | 2690 | struct btrfs_root *root = BTRFS_I(dir)->root; |
4df27c4d | 2691 | struct btrfs_path *path; |
a22285a6 | 2692 | struct btrfs_inode_ref *ref; |
4df27c4d | 2693 | struct btrfs_dir_item *di; |
7b128766 | 2694 | struct inode *inode = dentry->d_inode; |
4df27c4d | 2695 | u64 index; |
a22285a6 YZ |
2696 | int check_link = 1; |
2697 | int err = -ENOSPC; | |
4df27c4d | 2698 | int ret; |
33345d01 LZ |
2699 | u64 ino = btrfs_ino(inode); |
2700 | u64 dir_ino = btrfs_ino(dir); | |
4df27c4d | 2701 | |
e70bea5f JB |
2702 | /* |
2703 | * 1 for the possible orphan item | |
2704 | * 1 for the dir item | |
2705 | * 1 for the dir index | |
2706 | * 1 for the inode ref | |
2707 | * 1 for the inode ref in the tree log | |
2708 | * 2 for the dir entries in the log | |
2709 | * 1 for the inode | |
2710 | */ | |
2711 | trans = btrfs_start_transaction(root, 8); | |
a22285a6 YZ |
2712 | if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC) |
2713 | return trans; | |
4df27c4d | 2714 | |
33345d01 | 2715 | if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
a22285a6 | 2716 | return ERR_PTR(-ENOSPC); |
4df27c4d | 2717 | |
a22285a6 YZ |
2718 | /* check if there is someone else holds reference */ |
2719 | if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1) | |
2720 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2721 | |
a22285a6 YZ |
2722 | if (atomic_read(&inode->i_count) > 2) |
2723 | return ERR_PTR(-ENOSPC); | |
4df27c4d | 2724 | |
a22285a6 YZ |
2725 | if (xchg(&root->fs_info->enospc_unlink, 1)) |
2726 | return ERR_PTR(-ENOSPC); | |
2727 | ||
2728 | path = btrfs_alloc_path(); | |
2729 | if (!path) { | |
2730 | root->fs_info->enospc_unlink = 0; | |
2731 | return ERR_PTR(-ENOMEM); | |
4df27c4d YZ |
2732 | } |
2733 | ||
3880a1b4 JB |
2734 | /* 1 for the orphan item */ |
2735 | trans = btrfs_start_transaction(root, 1); | |
5df6a9f6 | 2736 | if (IS_ERR(trans)) { |
a22285a6 YZ |
2737 | btrfs_free_path(path); |
2738 | root->fs_info->enospc_unlink = 0; | |
2739 | return trans; | |
2740 | } | |
4df27c4d | 2741 | |
a22285a6 YZ |
2742 | path->skip_locking = 1; |
2743 | path->search_commit_root = 1; | |
4df27c4d | 2744 | |
a22285a6 YZ |
2745 | ret = btrfs_lookup_inode(trans, root, path, |
2746 | &BTRFS_I(dir)->location, 0); | |
2747 | if (ret < 0) { | |
2748 | err = ret; | |
2749 | goto out; | |
2750 | } | |
2751 | if (ret == 0) { | |
2752 | if (check_path_shared(root, path)) | |
2753 | goto out; | |
2754 | } else { | |
2755 | check_link = 0; | |
5df6a9f6 | 2756 | } |
b3b4aa74 | 2757 | btrfs_release_path(path); |
a22285a6 YZ |
2758 | |
2759 | ret = btrfs_lookup_inode(trans, root, path, | |
2760 | &BTRFS_I(inode)->location, 0); | |
2761 | if (ret < 0) { | |
2762 | err = ret; | |
2763 | goto out; | |
2764 | } | |
2765 | if (ret == 0) { | |
2766 | if (check_path_shared(root, path)) | |
2767 | goto out; | |
2768 | } else { | |
2769 | check_link = 0; | |
2770 | } | |
b3b4aa74 | 2771 | btrfs_release_path(path); |
a22285a6 YZ |
2772 | |
2773 | if (ret == 0 && S_ISREG(inode->i_mode)) { | |
2774 | ret = btrfs_lookup_file_extent(trans, root, path, | |
33345d01 | 2775 | ino, (u64)-1, 0); |
a22285a6 YZ |
2776 | if (ret < 0) { |
2777 | err = ret; | |
2778 | goto out; | |
2779 | } | |
2780 | BUG_ON(ret == 0); | |
2781 | if (check_path_shared(root, path)) | |
2782 | goto out; | |
b3b4aa74 | 2783 | btrfs_release_path(path); |
a22285a6 YZ |
2784 | } |
2785 | ||
2786 | if (!check_link) { | |
2787 | err = 0; | |
2788 | goto out; | |
2789 | } | |
2790 | ||
33345d01 | 2791 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
a22285a6 YZ |
2792 | dentry->d_name.name, dentry->d_name.len, 0); |
2793 | if (IS_ERR(di)) { | |
2794 | err = PTR_ERR(di); | |
2795 | goto out; | |
2796 | } | |
2797 | if (di) { | |
2798 | if (check_path_shared(root, path)) | |
2799 | goto out; | |
2800 | } else { | |
2801 | err = 0; | |
2802 | goto out; | |
2803 | } | |
b3b4aa74 | 2804 | btrfs_release_path(path); |
a22285a6 YZ |
2805 | |
2806 | ref = btrfs_lookup_inode_ref(trans, root, path, | |
2807 | dentry->d_name.name, dentry->d_name.len, | |
33345d01 | 2808 | ino, dir_ino, 0); |
a22285a6 YZ |
2809 | if (IS_ERR(ref)) { |
2810 | err = PTR_ERR(ref); | |
2811 | goto out; | |
2812 | } | |
2813 | BUG_ON(!ref); | |
2814 | if (check_path_shared(root, path)) | |
2815 | goto out; | |
2816 | index = btrfs_inode_ref_index(path->nodes[0], ref); | |
b3b4aa74 | 2817 | btrfs_release_path(path); |
a22285a6 | 2818 | |
16cdcec7 MX |
2819 | /* |
2820 | * This is a commit root search, if we can lookup inode item and other | |
2821 | * relative items in the commit root, it means the transaction of | |
2822 | * dir/file creation has been committed, and the dir index item that we | |
2823 | * delay to insert has also been inserted into the commit root. So | |
2824 | * we needn't worry about the delayed insertion of the dir index item | |
2825 | * here. | |
2826 | */ | |
33345d01 | 2827 | di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index, |
a22285a6 YZ |
2828 | dentry->d_name.name, dentry->d_name.len, 0); |
2829 | if (IS_ERR(di)) { | |
2830 | err = PTR_ERR(di); | |
2831 | goto out; | |
2832 | } | |
2833 | BUG_ON(ret == -ENOENT); | |
2834 | if (check_path_shared(root, path)) | |
2835 | goto out; | |
2836 | ||
2837 | err = 0; | |
2838 | out: | |
2839 | btrfs_free_path(path); | |
3880a1b4 JB |
2840 | /* Migrate the orphan reservation over */ |
2841 | if (!err) | |
2842 | err = btrfs_block_rsv_migrate(trans->block_rsv, | |
2843 | &root->fs_info->global_block_rsv, | |
5a77d76c | 2844 | trans->bytes_reserved); |
3880a1b4 | 2845 | |
a22285a6 YZ |
2846 | if (err) { |
2847 | btrfs_end_transaction(trans, root); | |
2848 | root->fs_info->enospc_unlink = 0; | |
2849 | return ERR_PTR(err); | |
2850 | } | |
2851 | ||
2852 | trans->block_rsv = &root->fs_info->global_block_rsv; | |
2853 | return trans; | |
2854 | } | |
2855 | ||
2856 | static void __unlink_end_trans(struct btrfs_trans_handle *trans, | |
2857 | struct btrfs_root *root) | |
2858 | { | |
2859 | if (trans->block_rsv == &root->fs_info->global_block_rsv) { | |
5a77d76c JB |
2860 | btrfs_block_rsv_release(root, trans->block_rsv, |
2861 | trans->bytes_reserved); | |
2862 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
a22285a6 YZ |
2863 | BUG_ON(!root->fs_info->enospc_unlink); |
2864 | root->fs_info->enospc_unlink = 0; | |
2865 | } | |
7ad85bb7 | 2866 | btrfs_end_transaction(trans, root); |
a22285a6 YZ |
2867 | } |
2868 | ||
2869 | static int btrfs_unlink(struct inode *dir, struct dentry *dentry) | |
2870 | { | |
2871 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
2872 | struct btrfs_trans_handle *trans; | |
2873 | struct inode *inode = dentry->d_inode; | |
2874 | int ret; | |
2875 | unsigned long nr = 0; | |
2876 | ||
2877 | trans = __unlink_start_trans(dir, dentry); | |
2878 | if (IS_ERR(trans)) | |
2879 | return PTR_ERR(trans); | |
5f39d397 | 2880 | |
12fcfd22 CM |
2881 | btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0); |
2882 | ||
e02119d5 CM |
2883 | ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2884 | dentry->d_name.name, dentry->d_name.len); | |
b532402e TI |
2885 | if (ret) |
2886 | goto out; | |
7b128766 | 2887 | |
a22285a6 | 2888 | if (inode->i_nlink == 0) { |
7b128766 | 2889 | ret = btrfs_orphan_add(trans, inode); |
b532402e TI |
2890 | if (ret) |
2891 | goto out; | |
a22285a6 | 2892 | } |
7b128766 | 2893 | |
b532402e | 2894 | out: |
d3c2fdcf | 2895 | nr = trans->blocks_used; |
a22285a6 | 2896 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2897 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
2898 | return ret; |
2899 | } | |
2900 | ||
4df27c4d YZ |
2901 | int btrfs_unlink_subvol(struct btrfs_trans_handle *trans, |
2902 | struct btrfs_root *root, | |
2903 | struct inode *dir, u64 objectid, | |
2904 | const char *name, int name_len) | |
2905 | { | |
2906 | struct btrfs_path *path; | |
2907 | struct extent_buffer *leaf; | |
2908 | struct btrfs_dir_item *di; | |
2909 | struct btrfs_key key; | |
2910 | u64 index; | |
2911 | int ret; | |
33345d01 | 2912 | u64 dir_ino = btrfs_ino(dir); |
4df27c4d YZ |
2913 | |
2914 | path = btrfs_alloc_path(); | |
2915 | if (!path) | |
2916 | return -ENOMEM; | |
2917 | ||
33345d01 | 2918 | di = btrfs_lookup_dir_item(trans, root, path, dir_ino, |
4df27c4d | 2919 | name, name_len, -1); |
c704005d | 2920 | BUG_ON(IS_ERR_OR_NULL(di)); |
4df27c4d YZ |
2921 | |
2922 | leaf = path->nodes[0]; | |
2923 | btrfs_dir_item_key_to_cpu(leaf, di, &key); | |
2924 | WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid); | |
2925 | ret = btrfs_delete_one_dir_name(trans, root, path, di); | |
2926 | BUG_ON(ret); | |
b3b4aa74 | 2927 | btrfs_release_path(path); |
4df27c4d YZ |
2928 | |
2929 | ret = btrfs_del_root_ref(trans, root->fs_info->tree_root, | |
2930 | objectid, root->root_key.objectid, | |
33345d01 | 2931 | dir_ino, &index, name, name_len); |
4df27c4d YZ |
2932 | if (ret < 0) { |
2933 | BUG_ON(ret != -ENOENT); | |
33345d01 | 2934 | di = btrfs_search_dir_index_item(root, path, dir_ino, |
4df27c4d | 2935 | name, name_len); |
c704005d | 2936 | BUG_ON(IS_ERR_OR_NULL(di)); |
4df27c4d YZ |
2937 | |
2938 | leaf = path->nodes[0]; | |
2939 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2940 | btrfs_release_path(path); |
4df27c4d YZ |
2941 | index = key.offset; |
2942 | } | |
945d8962 | 2943 | btrfs_release_path(path); |
4df27c4d | 2944 | |
16cdcec7 | 2945 | ret = btrfs_delete_delayed_dir_index(trans, root, dir, index); |
4df27c4d | 2946 | BUG_ON(ret); |
4df27c4d YZ |
2947 | |
2948 | btrfs_i_size_write(dir, dir->i_size - name_len * 2); | |
2949 | dir->i_mtime = dir->i_ctime = CURRENT_TIME; | |
2950 | ret = btrfs_update_inode(trans, root, dir); | |
2951 | BUG_ON(ret); | |
4df27c4d | 2952 | |
71d7aed0 | 2953 | btrfs_free_path(path); |
4df27c4d YZ |
2954 | return 0; |
2955 | } | |
2956 | ||
39279cc3 CM |
2957 | static int btrfs_rmdir(struct inode *dir, struct dentry *dentry) |
2958 | { | |
2959 | struct inode *inode = dentry->d_inode; | |
1832a6d5 | 2960 | int err = 0; |
39279cc3 | 2961 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 | 2962 | struct btrfs_trans_handle *trans; |
1832a6d5 | 2963 | unsigned long nr = 0; |
39279cc3 | 2964 | |
3394e160 | 2965 | if (inode->i_size > BTRFS_EMPTY_DIR_SIZE || |
33345d01 | 2966 | btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) |
134d4512 Y |
2967 | return -ENOTEMPTY; |
2968 | ||
a22285a6 YZ |
2969 | trans = __unlink_start_trans(dir, dentry); |
2970 | if (IS_ERR(trans)) | |
5df6a9f6 | 2971 | return PTR_ERR(trans); |
5df6a9f6 | 2972 | |
33345d01 | 2973 | if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
4df27c4d YZ |
2974 | err = btrfs_unlink_subvol(trans, root, dir, |
2975 | BTRFS_I(inode)->location.objectid, | |
2976 | dentry->d_name.name, | |
2977 | dentry->d_name.len); | |
2978 | goto out; | |
2979 | } | |
2980 | ||
7b128766 JB |
2981 | err = btrfs_orphan_add(trans, inode); |
2982 | if (err) | |
4df27c4d | 2983 | goto out; |
7b128766 | 2984 | |
39279cc3 | 2985 | /* now the directory is empty */ |
e02119d5 CM |
2986 | err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, |
2987 | dentry->d_name.name, dentry->d_name.len); | |
d397712b | 2988 | if (!err) |
dbe674a9 | 2989 | btrfs_i_size_write(inode, 0); |
4df27c4d | 2990 | out: |
d3c2fdcf | 2991 | nr = trans->blocks_used; |
a22285a6 | 2992 | __unlink_end_trans(trans, root); |
d3c2fdcf | 2993 | btrfs_btree_balance_dirty(root, nr); |
3954401f | 2994 | |
39279cc3 CM |
2995 | return err; |
2996 | } | |
2997 | ||
39279cc3 CM |
2998 | /* |
2999 | * this can truncate away extent items, csum items and directory items. | |
3000 | * It starts at a high offset and removes keys until it can't find | |
d352ac68 | 3001 | * any higher than new_size |
39279cc3 CM |
3002 | * |
3003 | * csum items that cross the new i_size are truncated to the new size | |
3004 | * as well. | |
7b128766 JB |
3005 | * |
3006 | * min_type is the minimum key type to truncate down to. If set to 0, this | |
3007 | * will kill all the items on this inode, including the INODE_ITEM_KEY. | |
39279cc3 | 3008 | */ |
8082510e YZ |
3009 | int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans, |
3010 | struct btrfs_root *root, | |
3011 | struct inode *inode, | |
3012 | u64 new_size, u32 min_type) | |
39279cc3 | 3013 | { |
39279cc3 | 3014 | struct btrfs_path *path; |
5f39d397 | 3015 | struct extent_buffer *leaf; |
39279cc3 | 3016 | struct btrfs_file_extent_item *fi; |
8082510e YZ |
3017 | struct btrfs_key key; |
3018 | struct btrfs_key found_key; | |
39279cc3 | 3019 | u64 extent_start = 0; |
db94535d | 3020 | u64 extent_num_bytes = 0; |
5d4f98a2 | 3021 | u64 extent_offset = 0; |
39279cc3 | 3022 | u64 item_end = 0; |
8082510e YZ |
3023 | u64 mask = root->sectorsize - 1; |
3024 | u32 found_type = (u8)-1; | |
39279cc3 CM |
3025 | int found_extent; |
3026 | int del_item; | |
85e21bac CM |
3027 | int pending_del_nr = 0; |
3028 | int pending_del_slot = 0; | |
179e29e4 | 3029 | int extent_type = -1; |
8082510e YZ |
3030 | int ret; |
3031 | int err = 0; | |
33345d01 | 3032 | u64 ino = btrfs_ino(inode); |
8082510e YZ |
3033 | |
3034 | BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY); | |
39279cc3 | 3035 | |
0eb0e19c MF |
3036 | path = btrfs_alloc_path(); |
3037 | if (!path) | |
3038 | return -ENOMEM; | |
3039 | path->reada = -1; | |
3040 | ||
0af3d00b | 3041 | if (root->ref_cows || root == root->fs_info->tree_root) |
5b21f2ed | 3042 | btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0); |
8082510e | 3043 | |
16cdcec7 MX |
3044 | /* |
3045 | * This function is also used to drop the items in the log tree before | |
3046 | * we relog the inode, so if root != BTRFS_I(inode)->root, it means | |
3047 | * it is used to drop the loged items. So we shouldn't kill the delayed | |
3048 | * items. | |
3049 | */ | |
3050 | if (min_type == 0 && root == BTRFS_I(inode)->root) | |
3051 | btrfs_kill_delayed_inode_items(inode); | |
3052 | ||
33345d01 | 3053 | key.objectid = ino; |
39279cc3 | 3054 | key.offset = (u64)-1; |
5f39d397 CM |
3055 | key.type = (u8)-1; |
3056 | ||
85e21bac | 3057 | search_again: |
b9473439 | 3058 | path->leave_spinning = 1; |
85e21bac | 3059 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
8082510e YZ |
3060 | if (ret < 0) { |
3061 | err = ret; | |
3062 | goto out; | |
3063 | } | |
d397712b | 3064 | |
85e21bac | 3065 | if (ret > 0) { |
e02119d5 CM |
3066 | /* there are no items in the tree for us to truncate, we're |
3067 | * done | |
3068 | */ | |
8082510e YZ |
3069 | if (path->slots[0] == 0) |
3070 | goto out; | |
85e21bac CM |
3071 | path->slots[0]--; |
3072 | } | |
3073 | ||
d397712b | 3074 | while (1) { |
39279cc3 | 3075 | fi = NULL; |
5f39d397 CM |
3076 | leaf = path->nodes[0]; |
3077 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
3078 | found_type = btrfs_key_type(&found_key); | |
39279cc3 | 3079 | |
33345d01 | 3080 | if (found_key.objectid != ino) |
39279cc3 | 3081 | break; |
5f39d397 | 3082 | |
85e21bac | 3083 | if (found_type < min_type) |
39279cc3 CM |
3084 | break; |
3085 | ||
5f39d397 | 3086 | item_end = found_key.offset; |
39279cc3 | 3087 | if (found_type == BTRFS_EXTENT_DATA_KEY) { |
5f39d397 | 3088 | fi = btrfs_item_ptr(leaf, path->slots[0], |
39279cc3 | 3089 | struct btrfs_file_extent_item); |
179e29e4 CM |
3090 | extent_type = btrfs_file_extent_type(leaf, fi); |
3091 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 3092 | item_end += |
db94535d | 3093 | btrfs_file_extent_num_bytes(leaf, fi); |
179e29e4 | 3094 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
179e29e4 | 3095 | item_end += btrfs_file_extent_inline_len(leaf, |
c8b97818 | 3096 | fi); |
39279cc3 | 3097 | } |
008630c1 | 3098 | item_end--; |
39279cc3 | 3099 | } |
8082510e YZ |
3100 | if (found_type > min_type) { |
3101 | del_item = 1; | |
3102 | } else { | |
3103 | if (item_end < new_size) | |
b888db2b | 3104 | break; |
8082510e YZ |
3105 | if (found_key.offset >= new_size) |
3106 | del_item = 1; | |
3107 | else | |
3108 | del_item = 0; | |
39279cc3 | 3109 | } |
39279cc3 | 3110 | found_extent = 0; |
39279cc3 | 3111 | /* FIXME, shrink the extent if the ref count is only 1 */ |
179e29e4 CM |
3112 | if (found_type != BTRFS_EXTENT_DATA_KEY) |
3113 | goto delete; | |
3114 | ||
3115 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
39279cc3 | 3116 | u64 num_dec; |
db94535d | 3117 | extent_start = btrfs_file_extent_disk_bytenr(leaf, fi); |
f70a9a6b | 3118 | if (!del_item) { |
db94535d CM |
3119 | u64 orig_num_bytes = |
3120 | btrfs_file_extent_num_bytes(leaf, fi); | |
e02119d5 | 3121 | extent_num_bytes = new_size - |
5f39d397 | 3122 | found_key.offset + root->sectorsize - 1; |
b1632b10 Y |
3123 | extent_num_bytes = extent_num_bytes & |
3124 | ~((u64)root->sectorsize - 1); | |
db94535d CM |
3125 | btrfs_set_file_extent_num_bytes(leaf, fi, |
3126 | extent_num_bytes); | |
3127 | num_dec = (orig_num_bytes - | |
9069218d | 3128 | extent_num_bytes); |
e02119d5 | 3129 | if (root->ref_cows && extent_start != 0) |
a76a3cd4 | 3130 | inode_sub_bytes(inode, num_dec); |
5f39d397 | 3131 | btrfs_mark_buffer_dirty(leaf); |
39279cc3 | 3132 | } else { |
db94535d CM |
3133 | extent_num_bytes = |
3134 | btrfs_file_extent_disk_num_bytes(leaf, | |
3135 | fi); | |
5d4f98a2 YZ |
3136 | extent_offset = found_key.offset - |
3137 | btrfs_file_extent_offset(leaf, fi); | |
3138 | ||
39279cc3 | 3139 | /* FIXME blocksize != 4096 */ |
9069218d | 3140 | num_dec = btrfs_file_extent_num_bytes(leaf, fi); |
39279cc3 CM |
3141 | if (extent_start != 0) { |
3142 | found_extent = 1; | |
e02119d5 | 3143 | if (root->ref_cows) |
a76a3cd4 | 3144 | inode_sub_bytes(inode, num_dec); |
e02119d5 | 3145 | } |
39279cc3 | 3146 | } |
9069218d | 3147 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
c8b97818 CM |
3148 | /* |
3149 | * we can't truncate inline items that have had | |
3150 | * special encodings | |
3151 | */ | |
3152 | if (!del_item && | |
3153 | btrfs_file_extent_compression(leaf, fi) == 0 && | |
3154 | btrfs_file_extent_encryption(leaf, fi) == 0 && | |
3155 | btrfs_file_extent_other_encoding(leaf, fi) == 0) { | |
e02119d5 CM |
3156 | u32 size = new_size - found_key.offset; |
3157 | ||
3158 | if (root->ref_cows) { | |
a76a3cd4 YZ |
3159 | inode_sub_bytes(inode, item_end + 1 - |
3160 | new_size); | |
e02119d5 CM |
3161 | } |
3162 | size = | |
3163 | btrfs_file_extent_calc_inline_size(size); | |
9069218d | 3164 | ret = btrfs_truncate_item(trans, root, path, |
e02119d5 | 3165 | size, 1); |
e02119d5 | 3166 | } else if (root->ref_cows) { |
a76a3cd4 YZ |
3167 | inode_sub_bytes(inode, item_end + 1 - |
3168 | found_key.offset); | |
9069218d | 3169 | } |
39279cc3 | 3170 | } |
179e29e4 | 3171 | delete: |
39279cc3 | 3172 | if (del_item) { |
85e21bac CM |
3173 | if (!pending_del_nr) { |
3174 | /* no pending yet, add ourselves */ | |
3175 | pending_del_slot = path->slots[0]; | |
3176 | pending_del_nr = 1; | |
3177 | } else if (pending_del_nr && | |
3178 | path->slots[0] + 1 == pending_del_slot) { | |
3179 | /* hop on the pending chunk */ | |
3180 | pending_del_nr++; | |
3181 | pending_del_slot = path->slots[0]; | |
3182 | } else { | |
d397712b | 3183 | BUG(); |
85e21bac | 3184 | } |
39279cc3 CM |
3185 | } else { |
3186 | break; | |
3187 | } | |
0af3d00b JB |
3188 | if (found_extent && (root->ref_cows || |
3189 | root == root->fs_info->tree_root)) { | |
b9473439 | 3190 | btrfs_set_path_blocking(path); |
39279cc3 | 3191 | ret = btrfs_free_extent(trans, root, extent_start, |
5d4f98a2 YZ |
3192 | extent_num_bytes, 0, |
3193 | btrfs_header_owner(leaf), | |
66d7e7f0 | 3194 | ino, extent_offset, 0); |
39279cc3 CM |
3195 | BUG_ON(ret); |
3196 | } | |
85e21bac | 3197 | |
8082510e YZ |
3198 | if (found_type == BTRFS_INODE_ITEM_KEY) |
3199 | break; | |
3200 | ||
3201 | if (path->slots[0] == 0 || | |
3202 | path->slots[0] != pending_del_slot) { | |
82d5902d LZ |
3203 | if (root->ref_cows && |
3204 | BTRFS_I(inode)->location.objectid != | |
3205 | BTRFS_FREE_INO_OBJECTID) { | |
8082510e YZ |
3206 | err = -EAGAIN; |
3207 | goto out; | |
3208 | } | |
3209 | if (pending_del_nr) { | |
3210 | ret = btrfs_del_items(trans, root, path, | |
3211 | pending_del_slot, | |
3212 | pending_del_nr); | |
3213 | BUG_ON(ret); | |
3214 | pending_del_nr = 0; | |
3215 | } | |
b3b4aa74 | 3216 | btrfs_release_path(path); |
85e21bac | 3217 | goto search_again; |
8082510e YZ |
3218 | } else { |
3219 | path->slots[0]--; | |
85e21bac | 3220 | } |
39279cc3 | 3221 | } |
8082510e | 3222 | out: |
85e21bac CM |
3223 | if (pending_del_nr) { |
3224 | ret = btrfs_del_items(trans, root, path, pending_del_slot, | |
3225 | pending_del_nr); | |
d68fc57b | 3226 | BUG_ON(ret); |
85e21bac | 3227 | } |
39279cc3 | 3228 | btrfs_free_path(path); |
8082510e | 3229 | return err; |
39279cc3 CM |
3230 | } |
3231 | ||
3232 | /* | |
3233 | * taken from block_truncate_page, but does cow as it zeros out | |
3234 | * any bytes left in the last page in the file. | |
3235 | */ | |
3236 | static int btrfs_truncate_page(struct address_space *mapping, loff_t from) | |
3237 | { | |
3238 | struct inode *inode = mapping->host; | |
db94535d | 3239 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
3240 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
3241 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 3242 | struct extent_state *cached_state = NULL; |
e6dcd2dc | 3243 | char *kaddr; |
db94535d | 3244 | u32 blocksize = root->sectorsize; |
39279cc3 CM |
3245 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
3246 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3247 | struct page *page; | |
3b16a4e3 | 3248 | gfp_t mask = btrfs_alloc_write_mask(mapping); |
39279cc3 | 3249 | int ret = 0; |
a52d9a80 | 3250 | u64 page_start; |
e6dcd2dc | 3251 | u64 page_end; |
39279cc3 CM |
3252 | |
3253 | if ((offset & (blocksize - 1)) == 0) | |
3254 | goto out; | |
0ca1f7ce | 3255 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
5d5e103a JB |
3256 | if (ret) |
3257 | goto out; | |
39279cc3 CM |
3258 | |
3259 | ret = -ENOMEM; | |
211c17f5 | 3260 | again: |
3b16a4e3 | 3261 | page = find_or_create_page(mapping, index, mask); |
5d5e103a | 3262 | if (!page) { |
0ca1f7ce | 3263 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 | 3264 | goto out; |
5d5e103a | 3265 | } |
e6dcd2dc CM |
3266 | |
3267 | page_start = page_offset(page); | |
3268 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
3269 | ||
39279cc3 | 3270 | if (!PageUptodate(page)) { |
9ebefb18 | 3271 | ret = btrfs_readpage(NULL, page); |
39279cc3 | 3272 | lock_page(page); |
211c17f5 CM |
3273 | if (page->mapping != mapping) { |
3274 | unlock_page(page); | |
3275 | page_cache_release(page); | |
3276 | goto again; | |
3277 | } | |
39279cc3 CM |
3278 | if (!PageUptodate(page)) { |
3279 | ret = -EIO; | |
89642229 | 3280 | goto out_unlock; |
39279cc3 CM |
3281 | } |
3282 | } | |
211c17f5 | 3283 | wait_on_page_writeback(page); |
e6dcd2dc | 3284 | |
2ac55d41 JB |
3285 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
3286 | GFP_NOFS); | |
e6dcd2dc CM |
3287 | set_page_extent_mapped(page); |
3288 | ||
3289 | ordered = btrfs_lookup_ordered_extent(inode, page_start); | |
3290 | if (ordered) { | |
2ac55d41 JB |
3291 | unlock_extent_cached(io_tree, page_start, page_end, |
3292 | &cached_state, GFP_NOFS); | |
e6dcd2dc CM |
3293 | unlock_page(page); |
3294 | page_cache_release(page); | |
eb84ae03 | 3295 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
3296 | btrfs_put_ordered_extent(ordered); |
3297 | goto again; | |
3298 | } | |
3299 | ||
2ac55d41 | 3300 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
5d5e103a | 3301 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 3302 | 0, 0, &cached_state, GFP_NOFS); |
5d5e103a | 3303 | |
2ac55d41 JB |
3304 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
3305 | &cached_state); | |
9ed74f2d | 3306 | if (ret) { |
2ac55d41 JB |
3307 | unlock_extent_cached(io_tree, page_start, page_end, |
3308 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
3309 | goto out_unlock; |
3310 | } | |
3311 | ||
e6dcd2dc CM |
3312 | ret = 0; |
3313 | if (offset != PAGE_CACHE_SIZE) { | |
3314 | kaddr = kmap(page); | |
3315 | memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); | |
3316 | flush_dcache_page(page); | |
3317 | kunmap(page); | |
3318 | } | |
247e743c | 3319 | ClearPageChecked(page); |
e6dcd2dc | 3320 | set_page_dirty(page); |
2ac55d41 JB |
3321 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, |
3322 | GFP_NOFS); | |
39279cc3 | 3323 | |
89642229 | 3324 | out_unlock: |
5d5e103a | 3325 | if (ret) |
0ca1f7ce | 3326 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
39279cc3 CM |
3327 | unlock_page(page); |
3328 | page_cache_release(page); | |
3329 | out: | |
3330 | return ret; | |
3331 | } | |
3332 | ||
695a0d0d JB |
3333 | /* |
3334 | * This function puts in dummy file extents for the area we're creating a hole | |
3335 | * for. So if we are truncating this file to a larger size we need to insert | |
3336 | * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for | |
3337 | * the range between oldsize and size | |
3338 | */ | |
a41ad394 | 3339 | int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size) |
39279cc3 | 3340 | { |
9036c102 YZ |
3341 | struct btrfs_trans_handle *trans; |
3342 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3343 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
a22285a6 | 3344 | struct extent_map *em = NULL; |
2ac55d41 | 3345 | struct extent_state *cached_state = NULL; |
9036c102 | 3346 | u64 mask = root->sectorsize - 1; |
a41ad394 | 3347 | u64 hole_start = (oldsize + mask) & ~mask; |
9036c102 YZ |
3348 | u64 block_end = (size + mask) & ~mask; |
3349 | u64 last_byte; | |
3350 | u64 cur_offset; | |
3351 | u64 hole_size; | |
9ed74f2d | 3352 | int err = 0; |
39279cc3 | 3353 | |
9036c102 YZ |
3354 | if (size <= hole_start) |
3355 | return 0; | |
3356 | ||
9036c102 YZ |
3357 | while (1) { |
3358 | struct btrfs_ordered_extent *ordered; | |
3359 | btrfs_wait_ordered_range(inode, hole_start, | |
3360 | block_end - hole_start); | |
2ac55d41 JB |
3361 | lock_extent_bits(io_tree, hole_start, block_end - 1, 0, |
3362 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3363 | ordered = btrfs_lookup_ordered_extent(inode, hole_start); |
3364 | if (!ordered) | |
3365 | break; | |
2ac55d41 JB |
3366 | unlock_extent_cached(io_tree, hole_start, block_end - 1, |
3367 | &cached_state, GFP_NOFS); | |
9036c102 YZ |
3368 | btrfs_put_ordered_extent(ordered); |
3369 | } | |
39279cc3 | 3370 | |
9036c102 YZ |
3371 | cur_offset = hole_start; |
3372 | while (1) { | |
3373 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, | |
3374 | block_end - cur_offset, 0); | |
c704005d | 3375 | BUG_ON(IS_ERR_OR_NULL(em)); |
9036c102 YZ |
3376 | last_byte = min(extent_map_end(em), block_end); |
3377 | last_byte = (last_byte + mask) & ~mask; | |
8082510e | 3378 | if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) { |
771ed689 | 3379 | u64 hint_byte = 0; |
9036c102 | 3380 | hole_size = last_byte - cur_offset; |
9ed74f2d | 3381 | |
3642320e | 3382 | trans = btrfs_start_transaction(root, 3); |
a22285a6 YZ |
3383 | if (IS_ERR(trans)) { |
3384 | err = PTR_ERR(trans); | |
9ed74f2d | 3385 | break; |
a22285a6 | 3386 | } |
8082510e YZ |
3387 | |
3388 | err = btrfs_drop_extents(trans, inode, cur_offset, | |
3389 | cur_offset + hole_size, | |
3390 | &hint_byte, 1); | |
5b397377 | 3391 | if (err) { |
3642320e | 3392 | btrfs_update_inode(trans, root, inode); |
5b397377 | 3393 | btrfs_end_transaction(trans, root); |
3893e33b | 3394 | break; |
5b397377 | 3395 | } |
8082510e | 3396 | |
9036c102 | 3397 | err = btrfs_insert_file_extent(trans, root, |
33345d01 | 3398 | btrfs_ino(inode), cur_offset, 0, |
9036c102 YZ |
3399 | 0, hole_size, 0, hole_size, |
3400 | 0, 0, 0); | |
5b397377 | 3401 | if (err) { |
3642320e | 3402 | btrfs_update_inode(trans, root, inode); |
5b397377 | 3403 | btrfs_end_transaction(trans, root); |
3893e33b | 3404 | break; |
5b397377 | 3405 | } |
8082510e | 3406 | |
9036c102 YZ |
3407 | btrfs_drop_extent_cache(inode, hole_start, |
3408 | last_byte - 1, 0); | |
8082510e | 3409 | |
3642320e | 3410 | btrfs_update_inode(trans, root, inode); |
8082510e | 3411 | btrfs_end_transaction(trans, root); |
9036c102 YZ |
3412 | } |
3413 | free_extent_map(em); | |
a22285a6 | 3414 | em = NULL; |
9036c102 | 3415 | cur_offset = last_byte; |
8082510e | 3416 | if (cur_offset >= block_end) |
9036c102 YZ |
3417 | break; |
3418 | } | |
1832a6d5 | 3419 | |
a22285a6 | 3420 | free_extent_map(em); |
2ac55d41 JB |
3421 | unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state, |
3422 | GFP_NOFS); | |
9036c102 YZ |
3423 | return err; |
3424 | } | |
39279cc3 | 3425 | |
a41ad394 | 3426 | static int btrfs_setsize(struct inode *inode, loff_t newsize) |
8082510e | 3427 | { |
f4a2f4c5 MX |
3428 | struct btrfs_root *root = BTRFS_I(inode)->root; |
3429 | struct btrfs_trans_handle *trans; | |
a41ad394 | 3430 | loff_t oldsize = i_size_read(inode); |
8082510e YZ |
3431 | int ret; |
3432 | ||
a41ad394 | 3433 | if (newsize == oldsize) |
8082510e YZ |
3434 | return 0; |
3435 | ||
a41ad394 | 3436 | if (newsize > oldsize) { |
a41ad394 JB |
3437 | truncate_pagecache(inode, oldsize, newsize); |
3438 | ret = btrfs_cont_expand(inode, oldsize, newsize); | |
f4a2f4c5 | 3439 | if (ret) |
8082510e | 3440 | return ret; |
8082510e | 3441 | |
f4a2f4c5 MX |
3442 | trans = btrfs_start_transaction(root, 1); |
3443 | if (IS_ERR(trans)) | |
3444 | return PTR_ERR(trans); | |
3445 | ||
3446 | i_size_write(inode, newsize); | |
3447 | btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL); | |
3448 | ret = btrfs_update_inode(trans, root, inode); | |
7ad85bb7 | 3449 | btrfs_end_transaction(trans, root); |
a41ad394 | 3450 | } else { |
8082510e | 3451 | |
a41ad394 JB |
3452 | /* |
3453 | * We're truncating a file that used to have good data down to | |
3454 | * zero. Make sure it gets into the ordered flush list so that | |
3455 | * any new writes get down to disk quickly. | |
3456 | */ | |
3457 | if (newsize == 0) | |
3458 | BTRFS_I(inode)->ordered_data_close = 1; | |
8082510e | 3459 | |
a41ad394 JB |
3460 | /* we don't support swapfiles, so vmtruncate shouldn't fail */ |
3461 | truncate_setsize(inode, newsize); | |
3462 | ret = btrfs_truncate(inode); | |
8082510e YZ |
3463 | } |
3464 | ||
a41ad394 | 3465 | return ret; |
8082510e YZ |
3466 | } |
3467 | ||
9036c102 YZ |
3468 | static int btrfs_setattr(struct dentry *dentry, struct iattr *attr) |
3469 | { | |
3470 | struct inode *inode = dentry->d_inode; | |
b83cc969 | 3471 | struct btrfs_root *root = BTRFS_I(inode)->root; |
9036c102 | 3472 | int err; |
39279cc3 | 3473 | |
b83cc969 LZ |
3474 | if (btrfs_root_readonly(root)) |
3475 | return -EROFS; | |
3476 | ||
9036c102 YZ |
3477 | err = inode_change_ok(inode, attr); |
3478 | if (err) | |
3479 | return err; | |
2bf5a725 | 3480 | |
5a3f23d5 | 3481 | if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { |
a41ad394 | 3482 | err = btrfs_setsize(inode, attr->ia_size); |
8082510e YZ |
3483 | if (err) |
3484 | return err; | |
39279cc3 | 3485 | } |
9036c102 | 3486 | |
1025774c CH |
3487 | if (attr->ia_valid) { |
3488 | setattr_copy(inode, attr); | |
22c44fe6 | 3489 | err = btrfs_dirty_inode(inode); |
1025774c | 3490 | |
22c44fe6 | 3491 | if (!err && attr->ia_valid & ATTR_MODE) |
1025774c CH |
3492 | err = btrfs_acl_chmod(inode); |
3493 | } | |
33268eaf | 3494 | |
39279cc3 CM |
3495 | return err; |
3496 | } | |
61295eb8 | 3497 | |
bd555975 | 3498 | void btrfs_evict_inode(struct inode *inode) |
39279cc3 CM |
3499 | { |
3500 | struct btrfs_trans_handle *trans; | |
3501 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
726c35fa | 3502 | struct btrfs_block_rsv *rsv, *global_rsv; |
07127184 | 3503 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
d3c2fdcf | 3504 | unsigned long nr; |
39279cc3 CM |
3505 | int ret; |
3506 | ||
1abe9b8a | 3507 | trace_btrfs_inode_evict(inode); |
3508 | ||
39279cc3 | 3509 | truncate_inode_pages(&inode->i_data, 0); |
0af3d00b | 3510 | if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 || |
2cf8572d | 3511 | btrfs_is_free_space_inode(root, inode))) |
bd555975 AV |
3512 | goto no_delete; |
3513 | ||
39279cc3 | 3514 | if (is_bad_inode(inode)) { |
7b128766 | 3515 | btrfs_orphan_del(NULL, inode); |
39279cc3 CM |
3516 | goto no_delete; |
3517 | } | |
bd555975 | 3518 | /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */ |
4a096752 | 3519 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
5f39d397 | 3520 | |
c71bf099 YZ |
3521 | if (root->fs_info->log_root_recovering) { |
3522 | BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan)); | |
3523 | goto no_delete; | |
3524 | } | |
3525 | ||
76dda93c YZ |
3526 | if (inode->i_nlink > 0) { |
3527 | BUG_ON(btrfs_root_refs(&root->root_item) != 0); | |
3528 | goto no_delete; | |
3529 | } | |
3530 | ||
4289a667 JB |
3531 | rsv = btrfs_alloc_block_rsv(root); |
3532 | if (!rsv) { | |
3533 | btrfs_orphan_del(NULL, inode); | |
3534 | goto no_delete; | |
3535 | } | |
4a338542 | 3536 | rsv->size = min_size; |
726c35fa | 3537 | global_rsv = &root->fs_info->global_block_rsv; |
4289a667 | 3538 | |
dbe674a9 | 3539 | btrfs_i_size_write(inode, 0); |
5f39d397 | 3540 | |
4289a667 JB |
3541 | /* |
3542 | * This is a bit simpler than btrfs_truncate since | |
3543 | * | |
3544 | * 1) We've already reserved our space for our orphan item in the | |
3545 | * unlink. | |
3546 | * 2) We're going to delete the inode item, so we don't need to update | |
3547 | * it at all. | |
3548 | * | |
3549 | * So we just need to reserve some slack space in case we add bytes when | |
3550 | * doing the truncate. | |
3551 | */ | |
8082510e | 3552 | while (1) { |
aa38a711 | 3553 | ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size); |
726c35fa JB |
3554 | |
3555 | /* | |
3556 | * Try and steal from the global reserve since we will | |
3557 | * likely not use this space anyway, we want to try as | |
3558 | * hard as possible to get this to work. | |
3559 | */ | |
3560 | if (ret) | |
3561 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size); | |
d68fc57b | 3562 | |
d68fc57b | 3563 | if (ret) { |
4289a667 | 3564 | printk(KERN_WARNING "Could not get space for a " |
482e6dc5 | 3565 | "delete, will truncate on mount %d\n", ret); |
4289a667 JB |
3566 | btrfs_orphan_del(NULL, inode); |
3567 | btrfs_free_block_rsv(root, rsv); | |
3568 | goto no_delete; | |
d68fc57b | 3569 | } |
7b128766 | 3570 | |
4289a667 JB |
3571 | trans = btrfs_start_transaction(root, 0); |
3572 | if (IS_ERR(trans)) { | |
3573 | btrfs_orphan_del(NULL, inode); | |
3574 | btrfs_free_block_rsv(root, rsv); | |
3575 | goto no_delete; | |
d68fc57b | 3576 | } |
7b128766 | 3577 | |
4289a667 JB |
3578 | trans->block_rsv = rsv; |
3579 | ||
d68fc57b | 3580 | ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0); |
8082510e YZ |
3581 | if (ret != -EAGAIN) |
3582 | break; | |
85e21bac | 3583 | |
8082510e YZ |
3584 | nr = trans->blocks_used; |
3585 | btrfs_end_transaction(trans, root); | |
3586 | trans = NULL; | |
3587 | btrfs_btree_balance_dirty(root, nr); | |
3588 | } | |
5f39d397 | 3589 | |
4289a667 JB |
3590 | btrfs_free_block_rsv(root, rsv); |
3591 | ||
8082510e | 3592 | if (ret == 0) { |
4289a667 | 3593 | trans->block_rsv = root->orphan_block_rsv; |
8082510e YZ |
3594 | ret = btrfs_orphan_del(trans, inode); |
3595 | BUG_ON(ret); | |
3596 | } | |
54aa1f4d | 3597 | |
4289a667 | 3598 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
581bb050 LZ |
3599 | if (!(root == root->fs_info->tree_root || |
3600 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)) | |
33345d01 | 3601 | btrfs_return_ino(root, btrfs_ino(inode)); |
581bb050 | 3602 | |
d3c2fdcf | 3603 | nr = trans->blocks_used; |
54aa1f4d | 3604 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 3605 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 | 3606 | no_delete: |
bd555975 | 3607 | end_writeback(inode); |
8082510e | 3608 | return; |
39279cc3 CM |
3609 | } |
3610 | ||
3611 | /* | |
3612 | * this returns the key found in the dir entry in the location pointer. | |
3613 | * If no dir entries were found, location->objectid is 0. | |
3614 | */ | |
3615 | static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry, | |
3616 | struct btrfs_key *location) | |
3617 | { | |
3618 | const char *name = dentry->d_name.name; | |
3619 | int namelen = dentry->d_name.len; | |
3620 | struct btrfs_dir_item *di; | |
3621 | struct btrfs_path *path; | |
3622 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
0d9f7f3e | 3623 | int ret = 0; |
39279cc3 CM |
3624 | |
3625 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
3626 | if (!path) |
3627 | return -ENOMEM; | |
3954401f | 3628 | |
33345d01 | 3629 | di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name, |
39279cc3 | 3630 | namelen, 0); |
0d9f7f3e Y |
3631 | if (IS_ERR(di)) |
3632 | ret = PTR_ERR(di); | |
d397712b | 3633 | |
c704005d | 3634 | if (IS_ERR_OR_NULL(di)) |
3954401f | 3635 | goto out_err; |
d397712b | 3636 | |
5f39d397 | 3637 | btrfs_dir_item_key_to_cpu(path->nodes[0], di, location); |
39279cc3 | 3638 | out: |
39279cc3 CM |
3639 | btrfs_free_path(path); |
3640 | return ret; | |
3954401f CM |
3641 | out_err: |
3642 | location->objectid = 0; | |
3643 | goto out; | |
39279cc3 CM |
3644 | } |
3645 | ||
3646 | /* | |
3647 | * when we hit a tree root in a directory, the btrfs part of the inode | |
3648 | * needs to be changed to reflect the root directory of the tree root. This | |
3649 | * is kind of like crossing a mount point. | |
3650 | */ | |
3651 | static int fixup_tree_root_location(struct btrfs_root *root, | |
4df27c4d YZ |
3652 | struct inode *dir, |
3653 | struct dentry *dentry, | |
3654 | struct btrfs_key *location, | |
3655 | struct btrfs_root **sub_root) | |
39279cc3 | 3656 | { |
4df27c4d YZ |
3657 | struct btrfs_path *path; |
3658 | struct btrfs_root *new_root; | |
3659 | struct btrfs_root_ref *ref; | |
3660 | struct extent_buffer *leaf; | |
3661 | int ret; | |
3662 | int err = 0; | |
39279cc3 | 3663 | |
4df27c4d YZ |
3664 | path = btrfs_alloc_path(); |
3665 | if (!path) { | |
3666 | err = -ENOMEM; | |
3667 | goto out; | |
3668 | } | |
39279cc3 | 3669 | |
4df27c4d YZ |
3670 | err = -ENOENT; |
3671 | ret = btrfs_find_root_ref(root->fs_info->tree_root, path, | |
3672 | BTRFS_I(dir)->root->root_key.objectid, | |
3673 | location->objectid); | |
3674 | if (ret) { | |
3675 | if (ret < 0) | |
3676 | err = ret; | |
3677 | goto out; | |
3678 | } | |
39279cc3 | 3679 | |
4df27c4d YZ |
3680 | leaf = path->nodes[0]; |
3681 | ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); | |
33345d01 | 3682 | if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) || |
4df27c4d YZ |
3683 | btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len) |
3684 | goto out; | |
39279cc3 | 3685 | |
4df27c4d YZ |
3686 | ret = memcmp_extent_buffer(leaf, dentry->d_name.name, |
3687 | (unsigned long)(ref + 1), | |
3688 | dentry->d_name.len); | |
3689 | if (ret) | |
3690 | goto out; | |
3691 | ||
b3b4aa74 | 3692 | btrfs_release_path(path); |
4df27c4d YZ |
3693 | |
3694 | new_root = btrfs_read_fs_root_no_name(root->fs_info, location); | |
3695 | if (IS_ERR(new_root)) { | |
3696 | err = PTR_ERR(new_root); | |
3697 | goto out; | |
3698 | } | |
3699 | ||
3700 | if (btrfs_root_refs(&new_root->root_item) == 0) { | |
3701 | err = -ENOENT; | |
3702 | goto out; | |
3703 | } | |
3704 | ||
3705 | *sub_root = new_root; | |
3706 | location->objectid = btrfs_root_dirid(&new_root->root_item); | |
3707 | location->type = BTRFS_INODE_ITEM_KEY; | |
3708 | location->offset = 0; | |
3709 | err = 0; | |
3710 | out: | |
3711 | btrfs_free_path(path); | |
3712 | return err; | |
39279cc3 CM |
3713 | } |
3714 | ||
5d4f98a2 YZ |
3715 | static void inode_tree_add(struct inode *inode) |
3716 | { | |
3717 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3718 | struct btrfs_inode *entry; | |
03e860bd FNP |
3719 | struct rb_node **p; |
3720 | struct rb_node *parent; | |
33345d01 | 3721 | u64 ino = btrfs_ino(inode); |
03e860bd FNP |
3722 | again: |
3723 | p = &root->inode_tree.rb_node; | |
3724 | parent = NULL; | |
5d4f98a2 | 3725 | |
1d3382cb | 3726 | if (inode_unhashed(inode)) |
76dda93c YZ |
3727 | return; |
3728 | ||
5d4f98a2 YZ |
3729 | spin_lock(&root->inode_lock); |
3730 | while (*p) { | |
3731 | parent = *p; | |
3732 | entry = rb_entry(parent, struct btrfs_inode, rb_node); | |
3733 | ||
33345d01 | 3734 | if (ino < btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3735 | p = &parent->rb_left; |
33345d01 | 3736 | else if (ino > btrfs_ino(&entry->vfs_inode)) |
03e860bd | 3737 | p = &parent->rb_right; |
5d4f98a2 YZ |
3738 | else { |
3739 | WARN_ON(!(entry->vfs_inode.i_state & | |
a4ffdde6 | 3740 | (I_WILL_FREE | I_FREEING))); |
03e860bd FNP |
3741 | rb_erase(parent, &root->inode_tree); |
3742 | RB_CLEAR_NODE(parent); | |
3743 | spin_unlock(&root->inode_lock); | |
3744 | goto again; | |
5d4f98a2 YZ |
3745 | } |
3746 | } | |
3747 | rb_link_node(&BTRFS_I(inode)->rb_node, parent, p); | |
3748 | rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree); | |
3749 | spin_unlock(&root->inode_lock); | |
3750 | } | |
3751 | ||
3752 | static void inode_tree_del(struct inode *inode) | |
3753 | { | |
3754 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
76dda93c | 3755 | int empty = 0; |
5d4f98a2 | 3756 | |
03e860bd | 3757 | spin_lock(&root->inode_lock); |
5d4f98a2 | 3758 | if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) { |
5d4f98a2 | 3759 | rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree); |
5d4f98a2 | 3760 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
76dda93c | 3761 | empty = RB_EMPTY_ROOT(&root->inode_tree); |
5d4f98a2 | 3762 | } |
03e860bd | 3763 | spin_unlock(&root->inode_lock); |
76dda93c | 3764 | |
0af3d00b JB |
3765 | /* |
3766 | * Free space cache has inodes in the tree root, but the tree root has a | |
3767 | * root_refs of 0, so this could end up dropping the tree root as a | |
3768 | * snapshot, so we need the extra !root->fs_info->tree_root check to | |
3769 | * make sure we don't drop it. | |
3770 | */ | |
3771 | if (empty && btrfs_root_refs(&root->root_item) == 0 && | |
3772 | root != root->fs_info->tree_root) { | |
76dda93c YZ |
3773 | synchronize_srcu(&root->fs_info->subvol_srcu); |
3774 | spin_lock(&root->inode_lock); | |
3775 | empty = RB_EMPTY_ROOT(&root->inode_tree); | |
3776 | spin_unlock(&root->inode_lock); | |
3777 | if (empty) | |
3778 | btrfs_add_dead_root(root); | |
3779 | } | |
3780 | } | |
3781 | ||
3782 | int btrfs_invalidate_inodes(struct btrfs_root *root) | |
3783 | { | |
3784 | struct rb_node *node; | |
3785 | struct rb_node *prev; | |
3786 | struct btrfs_inode *entry; | |
3787 | struct inode *inode; | |
3788 | u64 objectid = 0; | |
3789 | ||
3790 | WARN_ON(btrfs_root_refs(&root->root_item) != 0); | |
3791 | ||
3792 | spin_lock(&root->inode_lock); | |
3793 | again: | |
3794 | node = root->inode_tree.rb_node; | |
3795 | prev = NULL; | |
3796 | while (node) { | |
3797 | prev = node; | |
3798 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
3799 | ||
33345d01 | 3800 | if (objectid < btrfs_ino(&entry->vfs_inode)) |
76dda93c | 3801 | node = node->rb_left; |
33345d01 | 3802 | else if (objectid > btrfs_ino(&entry->vfs_inode)) |
76dda93c YZ |
3803 | node = node->rb_right; |
3804 | else | |
3805 | break; | |
3806 | } | |
3807 | if (!node) { | |
3808 | while (prev) { | |
3809 | entry = rb_entry(prev, struct btrfs_inode, rb_node); | |
33345d01 | 3810 | if (objectid <= btrfs_ino(&entry->vfs_inode)) { |
76dda93c YZ |
3811 | node = prev; |
3812 | break; | |
3813 | } | |
3814 | prev = rb_next(prev); | |
3815 | } | |
3816 | } | |
3817 | while (node) { | |
3818 | entry = rb_entry(node, struct btrfs_inode, rb_node); | |
33345d01 | 3819 | objectid = btrfs_ino(&entry->vfs_inode) + 1; |
76dda93c YZ |
3820 | inode = igrab(&entry->vfs_inode); |
3821 | if (inode) { | |
3822 | spin_unlock(&root->inode_lock); | |
3823 | if (atomic_read(&inode->i_count) > 1) | |
3824 | d_prune_aliases(inode); | |
3825 | /* | |
45321ac5 | 3826 | * btrfs_drop_inode will have it removed from |
76dda93c YZ |
3827 | * the inode cache when its usage count |
3828 | * hits zero. | |
3829 | */ | |
3830 | iput(inode); | |
3831 | cond_resched(); | |
3832 | spin_lock(&root->inode_lock); | |
3833 | goto again; | |
3834 | } | |
3835 | ||
3836 | if (cond_resched_lock(&root->inode_lock)) | |
3837 | goto again; | |
3838 | ||
3839 | node = rb_next(node); | |
3840 | } | |
3841 | spin_unlock(&root->inode_lock); | |
3842 | return 0; | |
5d4f98a2 YZ |
3843 | } |
3844 | ||
e02119d5 CM |
3845 | static int btrfs_init_locked_inode(struct inode *inode, void *p) |
3846 | { | |
3847 | struct btrfs_iget_args *args = p; | |
3848 | inode->i_ino = args->ino; | |
e02119d5 | 3849 | BTRFS_I(inode)->root = args->root; |
6a63209f | 3850 | btrfs_set_inode_space_info(args->root, inode); |
39279cc3 CM |
3851 | return 0; |
3852 | } | |
3853 | ||
3854 | static int btrfs_find_actor(struct inode *inode, void *opaque) | |
3855 | { | |
3856 | struct btrfs_iget_args *args = opaque; | |
33345d01 | 3857 | return args->ino == btrfs_ino(inode) && |
d397712b | 3858 | args->root == BTRFS_I(inode)->root; |
39279cc3 CM |
3859 | } |
3860 | ||
5d4f98a2 YZ |
3861 | static struct inode *btrfs_iget_locked(struct super_block *s, |
3862 | u64 objectid, | |
3863 | struct btrfs_root *root) | |
39279cc3 CM |
3864 | { |
3865 | struct inode *inode; | |
3866 | struct btrfs_iget_args args; | |
3867 | args.ino = objectid; | |
3868 | args.root = root; | |
3869 | ||
3870 | inode = iget5_locked(s, objectid, btrfs_find_actor, | |
3871 | btrfs_init_locked_inode, | |
3872 | (void *)&args); | |
3873 | return inode; | |
3874 | } | |
3875 | ||
1a54ef8c BR |
3876 | /* Get an inode object given its location and corresponding root. |
3877 | * Returns in *is_new if the inode was read from disk | |
3878 | */ | |
3879 | struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location, | |
73f73415 | 3880 | struct btrfs_root *root, int *new) |
1a54ef8c BR |
3881 | { |
3882 | struct inode *inode; | |
3883 | ||
3884 | inode = btrfs_iget_locked(s, location->objectid, root); | |
3885 | if (!inode) | |
5d4f98a2 | 3886 | return ERR_PTR(-ENOMEM); |
1a54ef8c BR |
3887 | |
3888 | if (inode->i_state & I_NEW) { | |
3889 | BTRFS_I(inode)->root = root; | |
3890 | memcpy(&BTRFS_I(inode)->location, location, sizeof(*location)); | |
3891 | btrfs_read_locked_inode(inode); | |
1748f843 MF |
3892 | if (!is_bad_inode(inode)) { |
3893 | inode_tree_add(inode); | |
3894 | unlock_new_inode(inode); | |
3895 | if (new) | |
3896 | *new = 1; | |
3897 | } else { | |
e0b6d65b ST |
3898 | unlock_new_inode(inode); |
3899 | iput(inode); | |
3900 | inode = ERR_PTR(-ESTALE); | |
1748f843 MF |
3901 | } |
3902 | } | |
3903 | ||
1a54ef8c BR |
3904 | return inode; |
3905 | } | |
3906 | ||
4df27c4d YZ |
3907 | static struct inode *new_simple_dir(struct super_block *s, |
3908 | struct btrfs_key *key, | |
3909 | struct btrfs_root *root) | |
3910 | { | |
3911 | struct inode *inode = new_inode(s); | |
3912 | ||
3913 | if (!inode) | |
3914 | return ERR_PTR(-ENOMEM); | |
3915 | ||
4df27c4d YZ |
3916 | BTRFS_I(inode)->root = root; |
3917 | memcpy(&BTRFS_I(inode)->location, key, sizeof(*key)); | |
3918 | BTRFS_I(inode)->dummy_inode = 1; | |
3919 | ||
3920 | inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID; | |
3921 | inode->i_op = &simple_dir_inode_operations; | |
3922 | inode->i_fop = &simple_dir_operations; | |
3923 | inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO; | |
3924 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; | |
3925 | ||
3926 | return inode; | |
3927 | } | |
3928 | ||
3de4586c | 3929 | struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry) |
39279cc3 | 3930 | { |
d397712b | 3931 | struct inode *inode; |
4df27c4d | 3932 | struct btrfs_root *root = BTRFS_I(dir)->root; |
39279cc3 CM |
3933 | struct btrfs_root *sub_root = root; |
3934 | struct btrfs_key location; | |
76dda93c | 3935 | int index; |
b4aff1f8 | 3936 | int ret = 0; |
39279cc3 CM |
3937 | |
3938 | if (dentry->d_name.len > BTRFS_NAME_LEN) | |
3939 | return ERR_PTR(-ENAMETOOLONG); | |
5f39d397 | 3940 | |
b4aff1f8 JB |
3941 | if (unlikely(d_need_lookup(dentry))) { |
3942 | memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key)); | |
3943 | kfree(dentry->d_fsdata); | |
3944 | dentry->d_fsdata = NULL; | |
a66e7cc6 JB |
3945 | /* This thing is hashed, drop it for now */ |
3946 | d_drop(dentry); | |
b4aff1f8 JB |
3947 | } else { |
3948 | ret = btrfs_inode_by_name(dir, dentry, &location); | |
3949 | } | |
5f39d397 | 3950 | |
39279cc3 CM |
3951 | if (ret < 0) |
3952 | return ERR_PTR(ret); | |
5f39d397 | 3953 | |
4df27c4d YZ |
3954 | if (location.objectid == 0) |
3955 | return NULL; | |
3956 | ||
3957 | if (location.type == BTRFS_INODE_ITEM_KEY) { | |
73f73415 | 3958 | inode = btrfs_iget(dir->i_sb, &location, root, NULL); |
4df27c4d YZ |
3959 | return inode; |
3960 | } | |
3961 | ||
3962 | BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY); | |
3963 | ||
76dda93c | 3964 | index = srcu_read_lock(&root->fs_info->subvol_srcu); |
4df27c4d YZ |
3965 | ret = fixup_tree_root_location(root, dir, dentry, |
3966 | &location, &sub_root); | |
3967 | if (ret < 0) { | |
3968 | if (ret != -ENOENT) | |
3969 | inode = ERR_PTR(ret); | |
3970 | else | |
3971 | inode = new_simple_dir(dir->i_sb, &location, sub_root); | |
3972 | } else { | |
73f73415 | 3973 | inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL); |
39279cc3 | 3974 | } |
76dda93c YZ |
3975 | srcu_read_unlock(&root->fs_info->subvol_srcu, index); |
3976 | ||
34d19bad | 3977 | if (!IS_ERR(inode) && root != sub_root) { |
c71bf099 YZ |
3978 | down_read(&root->fs_info->cleanup_work_sem); |
3979 | if (!(inode->i_sb->s_flags & MS_RDONLY)) | |
66b4ffd1 | 3980 | ret = btrfs_orphan_cleanup(sub_root); |
c71bf099 | 3981 | up_read(&root->fs_info->cleanup_work_sem); |
66b4ffd1 JB |
3982 | if (ret) |
3983 | inode = ERR_PTR(ret); | |
c71bf099 YZ |
3984 | } |
3985 | ||
3de4586c CM |
3986 | return inode; |
3987 | } | |
3988 | ||
fe15ce44 | 3989 | static int btrfs_dentry_delete(const struct dentry *dentry) |
76dda93c YZ |
3990 | { |
3991 | struct btrfs_root *root; | |
3992 | ||
efefb143 YZ |
3993 | if (!dentry->d_inode && !IS_ROOT(dentry)) |
3994 | dentry = dentry->d_parent; | |
76dda93c | 3995 | |
efefb143 YZ |
3996 | if (dentry->d_inode) { |
3997 | root = BTRFS_I(dentry->d_inode)->root; | |
3998 | if (btrfs_root_refs(&root->root_item) == 0) | |
3999 | return 1; | |
4000 | } | |
76dda93c YZ |
4001 | return 0; |
4002 | } | |
4003 | ||
b4aff1f8 JB |
4004 | static void btrfs_dentry_release(struct dentry *dentry) |
4005 | { | |
4006 | if (dentry->d_fsdata) | |
4007 | kfree(dentry->d_fsdata); | |
4008 | } | |
4009 | ||
3de4586c CM |
4010 | static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry, |
4011 | struct nameidata *nd) | |
4012 | { | |
a66e7cc6 JB |
4013 | struct dentry *ret; |
4014 | ||
4015 | ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry); | |
4016 | if (unlikely(d_need_lookup(dentry))) { | |
4017 | spin_lock(&dentry->d_lock); | |
4018 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
4019 | spin_unlock(&dentry->d_lock); | |
4020 | } | |
4021 | return ret; | |
39279cc3 CM |
4022 | } |
4023 | ||
16cdcec7 | 4024 | unsigned char btrfs_filetype_table[] = { |
39279cc3 CM |
4025 | DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK |
4026 | }; | |
4027 | ||
cbdf5a24 DW |
4028 | static int btrfs_real_readdir(struct file *filp, void *dirent, |
4029 | filldir_t filldir) | |
39279cc3 | 4030 | { |
6da6abae | 4031 | struct inode *inode = filp->f_dentry->d_inode; |
39279cc3 CM |
4032 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4033 | struct btrfs_item *item; | |
4034 | struct btrfs_dir_item *di; | |
4035 | struct btrfs_key key; | |
5f39d397 | 4036 | struct btrfs_key found_key; |
39279cc3 | 4037 | struct btrfs_path *path; |
16cdcec7 MX |
4038 | struct list_head ins_list; |
4039 | struct list_head del_list; | |
b4aff1f8 | 4040 | struct qstr q; |
39279cc3 | 4041 | int ret; |
5f39d397 | 4042 | struct extent_buffer *leaf; |
39279cc3 | 4043 | int slot; |
39279cc3 CM |
4044 | unsigned char d_type; |
4045 | int over = 0; | |
4046 | u32 di_cur; | |
4047 | u32 di_total; | |
4048 | u32 di_len; | |
4049 | int key_type = BTRFS_DIR_INDEX_KEY; | |
5f39d397 CM |
4050 | char tmp_name[32]; |
4051 | char *name_ptr; | |
4052 | int name_len; | |
16cdcec7 | 4053 | int is_curr = 0; /* filp->f_pos points to the current index? */ |
39279cc3 CM |
4054 | |
4055 | /* FIXME, use a real flag for deciding about the key type */ | |
4056 | if (root->fs_info->tree_root == root) | |
4057 | key_type = BTRFS_DIR_ITEM_KEY; | |
5f39d397 | 4058 | |
3954401f CM |
4059 | /* special case for "." */ |
4060 | if (filp->f_pos == 0) { | |
3765fefa HS |
4061 | over = filldir(dirent, ".", 1, |
4062 | filp->f_pos, btrfs_ino(inode), DT_DIR); | |
3954401f CM |
4063 | if (over) |
4064 | return 0; | |
4065 | filp->f_pos = 1; | |
4066 | } | |
3954401f CM |
4067 | /* special case for .., just use the back ref */ |
4068 | if (filp->f_pos == 1) { | |
5ecc7e5d | 4069 | u64 pino = parent_ino(filp->f_path.dentry); |
3954401f | 4070 | over = filldir(dirent, "..", 2, |
3765fefa | 4071 | filp->f_pos, pino, DT_DIR); |
3954401f | 4072 | if (over) |
49593bfa | 4073 | return 0; |
3954401f CM |
4074 | filp->f_pos = 2; |
4075 | } | |
49593bfa | 4076 | path = btrfs_alloc_path(); |
16cdcec7 MX |
4077 | if (!path) |
4078 | return -ENOMEM; | |
ff5714cc | 4079 | |
026fd317 | 4080 | path->reada = 1; |
49593bfa | 4081 | |
16cdcec7 MX |
4082 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4083 | INIT_LIST_HEAD(&ins_list); | |
4084 | INIT_LIST_HEAD(&del_list); | |
4085 | btrfs_get_delayed_items(inode, &ins_list, &del_list); | |
4086 | } | |
4087 | ||
39279cc3 CM |
4088 | btrfs_set_key_type(&key, key_type); |
4089 | key.offset = filp->f_pos; | |
33345d01 | 4090 | key.objectid = btrfs_ino(inode); |
5f39d397 | 4091 | |
39279cc3 CM |
4092 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4093 | if (ret < 0) | |
4094 | goto err; | |
49593bfa DW |
4095 | |
4096 | while (1) { | |
5f39d397 | 4097 | leaf = path->nodes[0]; |
39279cc3 | 4098 | slot = path->slots[0]; |
b9e03af0 LZ |
4099 | if (slot >= btrfs_header_nritems(leaf)) { |
4100 | ret = btrfs_next_leaf(root, path); | |
4101 | if (ret < 0) | |
4102 | goto err; | |
4103 | else if (ret > 0) | |
4104 | break; | |
4105 | continue; | |
39279cc3 | 4106 | } |
3de4586c | 4107 | |
5f39d397 CM |
4108 | item = btrfs_item_nr(leaf, slot); |
4109 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4110 | ||
4111 | if (found_key.objectid != key.objectid) | |
39279cc3 | 4112 | break; |
5f39d397 | 4113 | if (btrfs_key_type(&found_key) != key_type) |
39279cc3 | 4114 | break; |
5f39d397 | 4115 | if (found_key.offset < filp->f_pos) |
b9e03af0 | 4116 | goto next; |
16cdcec7 MX |
4117 | if (key_type == BTRFS_DIR_INDEX_KEY && |
4118 | btrfs_should_delete_dir_index(&del_list, | |
4119 | found_key.offset)) | |
4120 | goto next; | |
5f39d397 CM |
4121 | |
4122 | filp->f_pos = found_key.offset; | |
16cdcec7 | 4123 | is_curr = 1; |
49593bfa | 4124 | |
39279cc3 CM |
4125 | di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); |
4126 | di_cur = 0; | |
5f39d397 | 4127 | di_total = btrfs_item_size(leaf, item); |
49593bfa DW |
4128 | |
4129 | while (di_cur < di_total) { | |
5f39d397 | 4130 | struct btrfs_key location; |
b4aff1f8 | 4131 | struct dentry *tmp; |
5f39d397 | 4132 | |
22a94d44 JB |
4133 | if (verify_dir_item(root, leaf, di)) |
4134 | break; | |
4135 | ||
5f39d397 | 4136 | name_len = btrfs_dir_name_len(leaf, di); |
49593bfa | 4137 | if (name_len <= sizeof(tmp_name)) { |
5f39d397 CM |
4138 | name_ptr = tmp_name; |
4139 | } else { | |
4140 | name_ptr = kmalloc(name_len, GFP_NOFS); | |
49593bfa DW |
4141 | if (!name_ptr) { |
4142 | ret = -ENOMEM; | |
4143 | goto err; | |
4144 | } | |
5f39d397 CM |
4145 | } |
4146 | read_extent_buffer(leaf, name_ptr, | |
4147 | (unsigned long)(di + 1), name_len); | |
4148 | ||
4149 | d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)]; | |
4150 | btrfs_dir_item_key_to_cpu(leaf, di, &location); | |
3de4586c | 4151 | |
b4aff1f8 JB |
4152 | q.name = name_ptr; |
4153 | q.len = name_len; | |
4154 | q.hash = full_name_hash(q.name, q.len); | |
4155 | tmp = d_lookup(filp->f_dentry, &q); | |
4156 | if (!tmp) { | |
4157 | struct btrfs_key *newkey; | |
4158 | ||
4159 | newkey = kzalloc(sizeof(struct btrfs_key), | |
4160 | GFP_NOFS); | |
4161 | if (!newkey) | |
4162 | goto no_dentry; | |
4163 | tmp = d_alloc(filp->f_dentry, &q); | |
4164 | if (!tmp) { | |
4165 | kfree(newkey); | |
4166 | dput(tmp); | |
4167 | goto no_dentry; | |
4168 | } | |
4169 | memcpy(newkey, &location, | |
4170 | sizeof(struct btrfs_key)); | |
4171 | tmp->d_fsdata = newkey; | |
4172 | tmp->d_flags |= DCACHE_NEED_LOOKUP; | |
4173 | d_rehash(tmp); | |
4174 | dput(tmp); | |
4175 | } else { | |
4176 | dput(tmp); | |
4177 | } | |
4178 | no_dentry: | |
3de4586c CM |
4179 | /* is this a reference to our own snapshot? If so |
4180 | * skip it | |
4181 | */ | |
4182 | if (location.type == BTRFS_ROOT_ITEM_KEY && | |
4183 | location.objectid == root->root_key.objectid) { | |
4184 | over = 0; | |
4185 | goto skip; | |
4186 | } | |
5f39d397 | 4187 | over = filldir(dirent, name_ptr, name_len, |
49593bfa | 4188 | found_key.offset, location.objectid, |
39279cc3 | 4189 | d_type); |
5f39d397 | 4190 | |
3de4586c | 4191 | skip: |
5f39d397 CM |
4192 | if (name_ptr != tmp_name) |
4193 | kfree(name_ptr); | |
4194 | ||
39279cc3 CM |
4195 | if (over) |
4196 | goto nopos; | |
5103e947 | 4197 | di_len = btrfs_dir_name_len(leaf, di) + |
49593bfa | 4198 | btrfs_dir_data_len(leaf, di) + sizeof(*di); |
39279cc3 CM |
4199 | di_cur += di_len; |
4200 | di = (struct btrfs_dir_item *)((char *)di + di_len); | |
4201 | } | |
b9e03af0 LZ |
4202 | next: |
4203 | path->slots[0]++; | |
39279cc3 | 4204 | } |
49593bfa | 4205 | |
16cdcec7 MX |
4206 | if (key_type == BTRFS_DIR_INDEX_KEY) { |
4207 | if (is_curr) | |
4208 | filp->f_pos++; | |
4209 | ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir, | |
4210 | &ins_list); | |
4211 | if (ret) | |
4212 | goto nopos; | |
4213 | } | |
4214 | ||
49593bfa | 4215 | /* Reached end of directory/root. Bump pos past the last item. */ |
5e591a07 | 4216 | if (key_type == BTRFS_DIR_INDEX_KEY) |
406266ab JE |
4217 | /* |
4218 | * 32-bit glibc will use getdents64, but then strtol - | |
4219 | * so the last number we can serve is this. | |
4220 | */ | |
4221 | filp->f_pos = 0x7fffffff; | |
5e591a07 YZ |
4222 | else |
4223 | filp->f_pos++; | |
39279cc3 CM |
4224 | nopos: |
4225 | ret = 0; | |
4226 | err: | |
16cdcec7 MX |
4227 | if (key_type == BTRFS_DIR_INDEX_KEY) |
4228 | btrfs_put_delayed_items(&ins_list, &del_list); | |
39279cc3 | 4229 | btrfs_free_path(path); |
39279cc3 CM |
4230 | return ret; |
4231 | } | |
4232 | ||
a9185b41 | 4233 | int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc) |
39279cc3 CM |
4234 | { |
4235 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4236 | struct btrfs_trans_handle *trans; | |
4237 | int ret = 0; | |
0af3d00b | 4238 | bool nolock = false; |
39279cc3 | 4239 | |
8929ecfa | 4240 | if (BTRFS_I(inode)->dummy_inode) |
4ca8b41e CM |
4241 | return 0; |
4242 | ||
2cf8572d | 4243 | if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode)) |
82d5902d | 4244 | nolock = true; |
0af3d00b | 4245 | |
a9185b41 | 4246 | if (wbc->sync_mode == WB_SYNC_ALL) { |
0af3d00b | 4247 | if (nolock) |
7a7eaa40 | 4248 | trans = btrfs_join_transaction_nolock(root); |
0af3d00b | 4249 | else |
7a7eaa40 | 4250 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4251 | if (IS_ERR(trans)) |
4252 | return PTR_ERR(trans); | |
0af3d00b JB |
4253 | if (nolock) |
4254 | ret = btrfs_end_transaction_nolock(trans, root); | |
4255 | else | |
4256 | ret = btrfs_commit_transaction(trans, root); | |
39279cc3 CM |
4257 | } |
4258 | return ret; | |
4259 | } | |
4260 | ||
4261 | /* | |
54aa1f4d | 4262 | * This is somewhat expensive, updating the tree every time the |
39279cc3 CM |
4263 | * inode changes. But, it is most likely to find the inode in cache. |
4264 | * FIXME, needs more benchmarking...there are no reasons other than performance | |
4265 | * to keep or drop this code. | |
4266 | */ | |
22c44fe6 | 4267 | int btrfs_dirty_inode(struct inode *inode) |
39279cc3 CM |
4268 | { |
4269 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4270 | struct btrfs_trans_handle *trans; | |
8929ecfa YZ |
4271 | int ret; |
4272 | ||
4273 | if (BTRFS_I(inode)->dummy_inode) | |
22c44fe6 | 4274 | return 0; |
39279cc3 | 4275 | |
7a7eaa40 | 4276 | trans = btrfs_join_transaction(root); |
22c44fe6 JB |
4277 | if (IS_ERR(trans)) |
4278 | return PTR_ERR(trans); | |
8929ecfa YZ |
4279 | |
4280 | ret = btrfs_update_inode(trans, root, inode); | |
94b60442 CM |
4281 | if (ret && ret == -ENOSPC) { |
4282 | /* whoops, lets try again with the full transaction */ | |
4283 | btrfs_end_transaction(trans, root); | |
4284 | trans = btrfs_start_transaction(root, 1); | |
22c44fe6 JB |
4285 | if (IS_ERR(trans)) |
4286 | return PTR_ERR(trans); | |
8929ecfa | 4287 | |
94b60442 | 4288 | ret = btrfs_update_inode(trans, root, inode); |
94b60442 | 4289 | } |
39279cc3 | 4290 | btrfs_end_transaction(trans, root); |
16cdcec7 MX |
4291 | if (BTRFS_I(inode)->delayed_node) |
4292 | btrfs_balance_delayed_items(root); | |
22c44fe6 JB |
4293 | |
4294 | return ret; | |
4295 | } | |
4296 | ||
4297 | /* | |
4298 | * This is a copy of file_update_time. We need this so we can return error on | |
4299 | * ENOSPC for updating the inode in the case of file write and mmap writes. | |
4300 | */ | |
4301 | int btrfs_update_time(struct file *file) | |
4302 | { | |
4303 | struct inode *inode = file->f_path.dentry->d_inode; | |
4304 | struct timespec now; | |
4305 | int ret; | |
4306 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; | |
4307 | ||
4308 | /* First try to exhaust all avenues to not sync */ | |
4309 | if (IS_NOCMTIME(inode)) | |
4310 | return 0; | |
4311 | ||
4312 | now = current_fs_time(inode->i_sb); | |
4313 | if (!timespec_equal(&inode->i_mtime, &now)) | |
4314 | sync_it = S_MTIME; | |
4315 | ||
4316 | if (!timespec_equal(&inode->i_ctime, &now)) | |
4317 | sync_it |= S_CTIME; | |
4318 | ||
4319 | if (IS_I_VERSION(inode)) | |
4320 | sync_it |= S_VERSION; | |
4321 | ||
4322 | if (!sync_it) | |
4323 | return 0; | |
4324 | ||
4325 | /* Finally allowed to write? Takes lock. */ | |
4326 | if (mnt_want_write_file(file)) | |
4327 | return 0; | |
4328 | ||
4329 | /* Only change inode inside the lock region */ | |
4330 | if (sync_it & S_VERSION) | |
4331 | inode_inc_iversion(inode); | |
4332 | if (sync_it & S_CTIME) | |
4333 | inode->i_ctime = now; | |
4334 | if (sync_it & S_MTIME) | |
4335 | inode->i_mtime = now; | |
4336 | ret = btrfs_dirty_inode(inode); | |
4337 | if (!ret) | |
4338 | mark_inode_dirty_sync(inode); | |
4339 | mnt_drop_write(file->f_path.mnt); | |
4340 | return ret; | |
39279cc3 CM |
4341 | } |
4342 | ||
d352ac68 CM |
4343 | /* |
4344 | * find the highest existing sequence number in a directory | |
4345 | * and then set the in-memory index_cnt variable to reflect | |
4346 | * free sequence numbers | |
4347 | */ | |
aec7477b JB |
4348 | static int btrfs_set_inode_index_count(struct inode *inode) |
4349 | { | |
4350 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4351 | struct btrfs_key key, found_key; | |
4352 | struct btrfs_path *path; | |
4353 | struct extent_buffer *leaf; | |
4354 | int ret; | |
4355 | ||
33345d01 | 4356 | key.objectid = btrfs_ino(inode); |
aec7477b JB |
4357 | btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY); |
4358 | key.offset = (u64)-1; | |
4359 | ||
4360 | path = btrfs_alloc_path(); | |
4361 | if (!path) | |
4362 | return -ENOMEM; | |
4363 | ||
4364 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4365 | if (ret < 0) | |
4366 | goto out; | |
4367 | /* FIXME: we should be able to handle this */ | |
4368 | if (ret == 0) | |
4369 | goto out; | |
4370 | ret = 0; | |
4371 | ||
4372 | /* | |
4373 | * MAGIC NUMBER EXPLANATION: | |
4374 | * since we search a directory based on f_pos we have to start at 2 | |
4375 | * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody | |
4376 | * else has to start at 2 | |
4377 | */ | |
4378 | if (path->slots[0] == 0) { | |
4379 | BTRFS_I(inode)->index_cnt = 2; | |
4380 | goto out; | |
4381 | } | |
4382 | ||
4383 | path->slots[0]--; | |
4384 | ||
4385 | leaf = path->nodes[0]; | |
4386 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4387 | ||
33345d01 | 4388 | if (found_key.objectid != btrfs_ino(inode) || |
aec7477b JB |
4389 | btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) { |
4390 | BTRFS_I(inode)->index_cnt = 2; | |
4391 | goto out; | |
4392 | } | |
4393 | ||
4394 | BTRFS_I(inode)->index_cnt = found_key.offset + 1; | |
4395 | out: | |
4396 | btrfs_free_path(path); | |
4397 | return ret; | |
4398 | } | |
4399 | ||
d352ac68 CM |
4400 | /* |
4401 | * helper to find a free sequence number in a given directory. This current | |
4402 | * code is very simple, later versions will do smarter things in the btree | |
4403 | */ | |
3de4586c | 4404 | int btrfs_set_inode_index(struct inode *dir, u64 *index) |
aec7477b JB |
4405 | { |
4406 | int ret = 0; | |
4407 | ||
4408 | if (BTRFS_I(dir)->index_cnt == (u64)-1) { | |
16cdcec7 MX |
4409 | ret = btrfs_inode_delayed_dir_index_count(dir); |
4410 | if (ret) { | |
4411 | ret = btrfs_set_inode_index_count(dir); | |
4412 | if (ret) | |
4413 | return ret; | |
4414 | } | |
aec7477b JB |
4415 | } |
4416 | ||
00e4e6b3 | 4417 | *index = BTRFS_I(dir)->index_cnt; |
aec7477b JB |
4418 | BTRFS_I(dir)->index_cnt++; |
4419 | ||
4420 | return ret; | |
4421 | } | |
4422 | ||
39279cc3 CM |
4423 | static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans, |
4424 | struct btrfs_root *root, | |
aec7477b | 4425 | struct inode *dir, |
9c58309d | 4426 | const char *name, int name_len, |
d82a6f1d JB |
4427 | u64 ref_objectid, u64 objectid, int mode, |
4428 | u64 *index) | |
39279cc3 CM |
4429 | { |
4430 | struct inode *inode; | |
5f39d397 | 4431 | struct btrfs_inode_item *inode_item; |
39279cc3 | 4432 | struct btrfs_key *location; |
5f39d397 | 4433 | struct btrfs_path *path; |
9c58309d CM |
4434 | struct btrfs_inode_ref *ref; |
4435 | struct btrfs_key key[2]; | |
4436 | u32 sizes[2]; | |
4437 | unsigned long ptr; | |
39279cc3 CM |
4438 | int ret; |
4439 | int owner; | |
4440 | ||
5f39d397 | 4441 | path = btrfs_alloc_path(); |
d8926bb3 MF |
4442 | if (!path) |
4443 | return ERR_PTR(-ENOMEM); | |
5f39d397 | 4444 | |
39279cc3 | 4445 | inode = new_inode(root->fs_info->sb); |
8fb27640 YS |
4446 | if (!inode) { |
4447 | btrfs_free_path(path); | |
39279cc3 | 4448 | return ERR_PTR(-ENOMEM); |
8fb27640 | 4449 | } |
39279cc3 | 4450 | |
581bb050 LZ |
4451 | /* |
4452 | * we have to initialize this early, so we can reclaim the inode | |
4453 | * number if we fail afterwards in this function. | |
4454 | */ | |
4455 | inode->i_ino = objectid; | |
4456 | ||
aec7477b | 4457 | if (dir) { |
1abe9b8a | 4458 | trace_btrfs_inode_request(dir); |
4459 | ||
3de4586c | 4460 | ret = btrfs_set_inode_index(dir, index); |
09771430 | 4461 | if (ret) { |
8fb27640 | 4462 | btrfs_free_path(path); |
09771430 | 4463 | iput(inode); |
aec7477b | 4464 | return ERR_PTR(ret); |
09771430 | 4465 | } |
aec7477b JB |
4466 | } |
4467 | /* | |
4468 | * index_cnt is ignored for everything but a dir, | |
4469 | * btrfs_get_inode_index_count has an explanation for the magic | |
4470 | * number | |
4471 | */ | |
4472 | BTRFS_I(inode)->index_cnt = 2; | |
39279cc3 | 4473 | BTRFS_I(inode)->root = root; |
e02119d5 | 4474 | BTRFS_I(inode)->generation = trans->transid; |
76195853 | 4475 | inode->i_generation = BTRFS_I(inode)->generation; |
6a63209f | 4476 | btrfs_set_inode_space_info(root, inode); |
b888db2b | 4477 | |
569254b0 | 4478 | if (S_ISDIR(mode)) |
39279cc3 CM |
4479 | owner = 0; |
4480 | else | |
4481 | owner = 1; | |
9c58309d CM |
4482 | |
4483 | key[0].objectid = objectid; | |
4484 | btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); | |
4485 | key[0].offset = 0; | |
4486 | ||
4487 | key[1].objectid = objectid; | |
4488 | btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY); | |
4489 | key[1].offset = ref_objectid; | |
4490 | ||
4491 | sizes[0] = sizeof(struct btrfs_inode_item); | |
4492 | sizes[1] = name_len + sizeof(*ref); | |
4493 | ||
b9473439 | 4494 | path->leave_spinning = 1; |
9c58309d CM |
4495 | ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); |
4496 | if (ret != 0) | |
5f39d397 CM |
4497 | goto fail; |
4498 | ||
ecc11fab | 4499 | inode_init_owner(inode, dir, mode); |
a76a3cd4 | 4500 | inode_set_bytes(inode, 0); |
39279cc3 | 4501 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; |
5f39d397 CM |
4502 | inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], |
4503 | struct btrfs_inode_item); | |
e02119d5 | 4504 | fill_inode_item(trans, path->nodes[0], inode_item, inode); |
9c58309d CM |
4505 | |
4506 | ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, | |
4507 | struct btrfs_inode_ref); | |
4508 | btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len); | |
00e4e6b3 | 4509 | btrfs_set_inode_ref_index(path->nodes[0], ref, *index); |
9c58309d CM |
4510 | ptr = (unsigned long)(ref + 1); |
4511 | write_extent_buffer(path->nodes[0], name, ptr, name_len); | |
4512 | ||
5f39d397 CM |
4513 | btrfs_mark_buffer_dirty(path->nodes[0]); |
4514 | btrfs_free_path(path); | |
4515 | ||
39279cc3 CM |
4516 | location = &BTRFS_I(inode)->location; |
4517 | location->objectid = objectid; | |
39279cc3 CM |
4518 | location->offset = 0; |
4519 | btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY); | |
4520 | ||
6cbff00f CH |
4521 | btrfs_inherit_iflags(inode, dir); |
4522 | ||
569254b0 | 4523 | if (S_ISREG(mode)) { |
94272164 CM |
4524 | if (btrfs_test_opt(root, NODATASUM)) |
4525 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; | |
75e7cb7f LB |
4526 | if (btrfs_test_opt(root, NODATACOW) || |
4527 | (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW)) | |
94272164 CM |
4528 | BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; |
4529 | } | |
4530 | ||
39279cc3 | 4531 | insert_inode_hash(inode); |
5d4f98a2 | 4532 | inode_tree_add(inode); |
1abe9b8a | 4533 | |
4534 | trace_btrfs_inode_new(inode); | |
1973f0fa | 4535 | btrfs_set_inode_last_trans(trans, inode); |
1abe9b8a | 4536 | |
39279cc3 | 4537 | return inode; |
5f39d397 | 4538 | fail: |
aec7477b JB |
4539 | if (dir) |
4540 | BTRFS_I(dir)->index_cnt--; | |
5f39d397 | 4541 | btrfs_free_path(path); |
09771430 | 4542 | iput(inode); |
5f39d397 | 4543 | return ERR_PTR(ret); |
39279cc3 CM |
4544 | } |
4545 | ||
4546 | static inline u8 btrfs_inode_type(struct inode *inode) | |
4547 | { | |
4548 | return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; | |
4549 | } | |
4550 | ||
d352ac68 CM |
4551 | /* |
4552 | * utility function to add 'inode' into 'parent_inode' with | |
4553 | * a give name and a given sequence number. | |
4554 | * if 'add_backref' is true, also insert a backref from the | |
4555 | * inode to the parent directory. | |
4556 | */ | |
e02119d5 CM |
4557 | int btrfs_add_link(struct btrfs_trans_handle *trans, |
4558 | struct inode *parent_inode, struct inode *inode, | |
4559 | const char *name, int name_len, int add_backref, u64 index) | |
39279cc3 | 4560 | { |
4df27c4d | 4561 | int ret = 0; |
39279cc3 | 4562 | struct btrfs_key key; |
e02119d5 | 4563 | struct btrfs_root *root = BTRFS_I(parent_inode)->root; |
33345d01 LZ |
4564 | u64 ino = btrfs_ino(inode); |
4565 | u64 parent_ino = btrfs_ino(parent_inode); | |
5f39d397 | 4566 | |
33345d01 | 4567 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4568 | memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key)); |
4569 | } else { | |
33345d01 | 4570 | key.objectid = ino; |
4df27c4d YZ |
4571 | btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); |
4572 | key.offset = 0; | |
4573 | } | |
4574 | ||
33345d01 | 4575 | if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
4576 | ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, |
4577 | key.objectid, root->root_key.objectid, | |
33345d01 | 4578 | parent_ino, index, name, name_len); |
4df27c4d | 4579 | } else if (add_backref) { |
33345d01 LZ |
4580 | ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino, |
4581 | parent_ino, index); | |
4df27c4d | 4582 | } |
39279cc3 | 4583 | |
39279cc3 | 4584 | if (ret == 0) { |
4df27c4d | 4585 | ret = btrfs_insert_dir_item(trans, root, name, name_len, |
16cdcec7 | 4586 | parent_inode, &key, |
4df27c4d YZ |
4587 | btrfs_inode_type(inode), index); |
4588 | BUG_ON(ret); | |
4589 | ||
dbe674a9 | 4590 | btrfs_i_size_write(parent_inode, parent_inode->i_size + |
e02119d5 | 4591 | name_len * 2); |
79c44584 | 4592 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; |
e02119d5 | 4593 | ret = btrfs_update_inode(trans, root, parent_inode); |
39279cc3 CM |
4594 | } |
4595 | return ret; | |
4596 | } | |
4597 | ||
4598 | static int btrfs_add_nondir(struct btrfs_trans_handle *trans, | |
a1b075d2 JB |
4599 | struct inode *dir, struct dentry *dentry, |
4600 | struct inode *inode, int backref, u64 index) | |
39279cc3 | 4601 | { |
a1b075d2 JB |
4602 | int err = btrfs_add_link(trans, dir, inode, |
4603 | dentry->d_name.name, dentry->d_name.len, | |
4604 | backref, index); | |
39279cc3 CM |
4605 | if (err > 0) |
4606 | err = -EEXIST; | |
4607 | return err; | |
4608 | } | |
4609 | ||
618e21d5 JB |
4610 | static int btrfs_mknod(struct inode *dir, struct dentry *dentry, |
4611 | int mode, dev_t rdev) | |
4612 | { | |
4613 | struct btrfs_trans_handle *trans; | |
4614 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4615 | struct inode *inode = NULL; |
618e21d5 JB |
4616 | int err; |
4617 | int drop_inode = 0; | |
4618 | u64 objectid; | |
1832a6d5 | 4619 | unsigned long nr = 0; |
00e4e6b3 | 4620 | u64 index = 0; |
618e21d5 JB |
4621 | |
4622 | if (!new_valid_dev(rdev)) | |
4623 | return -EINVAL; | |
4624 | ||
9ed74f2d JB |
4625 | /* |
4626 | * 2 for inode item and ref | |
4627 | * 2 for dir items | |
4628 | * 1 for xattr if selinux is on | |
4629 | */ | |
a22285a6 YZ |
4630 | trans = btrfs_start_transaction(root, 5); |
4631 | if (IS_ERR(trans)) | |
4632 | return PTR_ERR(trans); | |
1832a6d5 | 4633 | |
581bb050 LZ |
4634 | err = btrfs_find_free_ino(root, &objectid); |
4635 | if (err) | |
4636 | goto out_unlock; | |
4637 | ||
aec7477b | 4638 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4639 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4640 | mode, &index); |
7cf96da3 TI |
4641 | if (IS_ERR(inode)) { |
4642 | err = PTR_ERR(inode); | |
618e21d5 | 4643 | goto out_unlock; |
7cf96da3 | 4644 | } |
618e21d5 | 4645 | |
2a7dba39 | 4646 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4647 | if (err) { |
4648 | drop_inode = 1; | |
4649 | goto out_unlock; | |
4650 | } | |
4651 | ||
ad19db71 CS |
4652 | /* |
4653 | * If the active LSM wants to access the inode during | |
4654 | * d_instantiate it needs these. Smack checks to see | |
4655 | * if the filesystem supports xattrs by looking at the | |
4656 | * ops vector. | |
4657 | */ | |
4658 | ||
4659 | inode->i_op = &btrfs_special_inode_operations; | |
a1b075d2 | 4660 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
618e21d5 JB |
4661 | if (err) |
4662 | drop_inode = 1; | |
4663 | else { | |
618e21d5 | 4664 | init_special_inode(inode, inode->i_mode, rdev); |
1b4ab1bb | 4665 | btrfs_update_inode(trans, root, inode); |
08c422c2 | 4666 | d_instantiate(dentry, inode); |
618e21d5 | 4667 | } |
618e21d5 | 4668 | out_unlock: |
d3c2fdcf | 4669 | nr = trans->blocks_used; |
7ad85bb7 | 4670 | btrfs_end_transaction(trans, root); |
a22285a6 | 4671 | btrfs_btree_balance_dirty(root, nr); |
618e21d5 JB |
4672 | if (drop_inode) { |
4673 | inode_dec_link_count(inode); | |
4674 | iput(inode); | |
4675 | } | |
618e21d5 JB |
4676 | return err; |
4677 | } | |
4678 | ||
39279cc3 CM |
4679 | static int btrfs_create(struct inode *dir, struct dentry *dentry, |
4680 | int mode, struct nameidata *nd) | |
4681 | { | |
4682 | struct btrfs_trans_handle *trans; | |
4683 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
1832a6d5 | 4684 | struct inode *inode = NULL; |
39279cc3 | 4685 | int drop_inode = 0; |
a22285a6 | 4686 | int err; |
1832a6d5 | 4687 | unsigned long nr = 0; |
39279cc3 | 4688 | u64 objectid; |
00e4e6b3 | 4689 | u64 index = 0; |
39279cc3 | 4690 | |
9ed74f2d JB |
4691 | /* |
4692 | * 2 for inode item and ref | |
4693 | * 2 for dir items | |
4694 | * 1 for xattr if selinux is on | |
4695 | */ | |
a22285a6 YZ |
4696 | trans = btrfs_start_transaction(root, 5); |
4697 | if (IS_ERR(trans)) | |
4698 | return PTR_ERR(trans); | |
9ed74f2d | 4699 | |
581bb050 LZ |
4700 | err = btrfs_find_free_ino(root, &objectid); |
4701 | if (err) | |
4702 | goto out_unlock; | |
4703 | ||
aec7477b | 4704 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4705 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4706 | mode, &index); |
7cf96da3 TI |
4707 | if (IS_ERR(inode)) { |
4708 | err = PTR_ERR(inode); | |
39279cc3 | 4709 | goto out_unlock; |
7cf96da3 | 4710 | } |
39279cc3 | 4711 | |
2a7dba39 | 4712 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4713 | if (err) { |
4714 | drop_inode = 1; | |
4715 | goto out_unlock; | |
4716 | } | |
4717 | ||
ad19db71 CS |
4718 | /* |
4719 | * If the active LSM wants to access the inode during | |
4720 | * d_instantiate it needs these. Smack checks to see | |
4721 | * if the filesystem supports xattrs by looking at the | |
4722 | * ops vector. | |
4723 | */ | |
4724 | inode->i_fop = &btrfs_file_operations; | |
4725 | inode->i_op = &btrfs_file_inode_operations; | |
4726 | ||
a1b075d2 | 4727 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
4728 | if (err) |
4729 | drop_inode = 1; | |
4730 | else { | |
4731 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 4732 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 4733 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
08c422c2 | 4734 | d_instantiate(dentry, inode); |
39279cc3 | 4735 | } |
39279cc3 | 4736 | out_unlock: |
d3c2fdcf | 4737 | nr = trans->blocks_used; |
7ad85bb7 | 4738 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4739 | if (drop_inode) { |
4740 | inode_dec_link_count(inode); | |
4741 | iput(inode); | |
4742 | } | |
d3c2fdcf | 4743 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4744 | return err; |
4745 | } | |
4746 | ||
4747 | static int btrfs_link(struct dentry *old_dentry, struct inode *dir, | |
4748 | struct dentry *dentry) | |
4749 | { | |
4750 | struct btrfs_trans_handle *trans; | |
4751 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4752 | struct inode *inode = old_dentry->d_inode; | |
00e4e6b3 | 4753 | u64 index; |
1832a6d5 | 4754 | unsigned long nr = 0; |
39279cc3 CM |
4755 | int err; |
4756 | int drop_inode = 0; | |
4757 | ||
4a8be425 TH |
4758 | /* do not allow sys_link's with other subvols of the same device */ |
4759 | if (root->objectid != BTRFS_I(inode)->root->objectid) | |
3ab3564f | 4760 | return -EXDEV; |
4a8be425 | 4761 | |
c055e99e AV |
4762 | if (inode->i_nlink == ~0U) |
4763 | return -EMLINK; | |
4a8be425 | 4764 | |
3de4586c | 4765 | err = btrfs_set_inode_index(dir, &index); |
aec7477b JB |
4766 | if (err) |
4767 | goto fail; | |
4768 | ||
a22285a6 | 4769 | /* |
7e6b6465 | 4770 | * 2 items for inode and inode ref |
a22285a6 | 4771 | * 2 items for dir items |
7e6b6465 | 4772 | * 1 item for parent inode |
a22285a6 | 4773 | */ |
7e6b6465 | 4774 | trans = btrfs_start_transaction(root, 5); |
a22285a6 YZ |
4775 | if (IS_ERR(trans)) { |
4776 | err = PTR_ERR(trans); | |
4777 | goto fail; | |
4778 | } | |
5f39d397 | 4779 | |
3153495d MX |
4780 | btrfs_inc_nlink(inode); |
4781 | inode->i_ctime = CURRENT_TIME; | |
7de9c6ee | 4782 | ihold(inode); |
aec7477b | 4783 | |
a1b075d2 | 4784 | err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index); |
5f39d397 | 4785 | |
a5719521 | 4786 | if (err) { |
54aa1f4d | 4787 | drop_inode = 1; |
a5719521 | 4788 | } else { |
10d9f309 | 4789 | struct dentry *parent = dentry->d_parent; |
a5719521 YZ |
4790 | err = btrfs_update_inode(trans, root, inode); |
4791 | BUG_ON(err); | |
08c422c2 | 4792 | d_instantiate(dentry, inode); |
6a912213 | 4793 | btrfs_log_new_name(trans, inode, NULL, parent); |
a5719521 | 4794 | } |
39279cc3 | 4795 | |
d3c2fdcf | 4796 | nr = trans->blocks_used; |
7ad85bb7 | 4797 | btrfs_end_transaction(trans, root); |
1832a6d5 | 4798 | fail: |
39279cc3 CM |
4799 | if (drop_inode) { |
4800 | inode_dec_link_count(inode); | |
4801 | iput(inode); | |
4802 | } | |
d3c2fdcf | 4803 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4804 | return err; |
4805 | } | |
4806 | ||
39279cc3 CM |
4807 | static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) |
4808 | { | |
b9d86667 | 4809 | struct inode *inode = NULL; |
39279cc3 CM |
4810 | struct btrfs_trans_handle *trans; |
4811 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
4812 | int err = 0; | |
4813 | int drop_on_err = 0; | |
b9d86667 | 4814 | u64 objectid = 0; |
00e4e6b3 | 4815 | u64 index = 0; |
d3c2fdcf | 4816 | unsigned long nr = 1; |
39279cc3 | 4817 | |
9ed74f2d JB |
4818 | /* |
4819 | * 2 items for inode and ref | |
4820 | * 2 items for dir items | |
4821 | * 1 for xattr if selinux is on | |
4822 | */ | |
a22285a6 YZ |
4823 | trans = btrfs_start_transaction(root, 5); |
4824 | if (IS_ERR(trans)) | |
4825 | return PTR_ERR(trans); | |
39279cc3 | 4826 | |
581bb050 LZ |
4827 | err = btrfs_find_free_ino(root, &objectid); |
4828 | if (err) | |
4829 | goto out_fail; | |
4830 | ||
aec7477b | 4831 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 4832 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 4833 | S_IFDIR | mode, &index); |
39279cc3 CM |
4834 | if (IS_ERR(inode)) { |
4835 | err = PTR_ERR(inode); | |
4836 | goto out_fail; | |
4837 | } | |
5f39d397 | 4838 | |
39279cc3 | 4839 | drop_on_err = 1; |
33268eaf | 4840 | |
2a7dba39 | 4841 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
4842 | if (err) |
4843 | goto out_fail; | |
4844 | ||
39279cc3 CM |
4845 | inode->i_op = &btrfs_dir_inode_operations; |
4846 | inode->i_fop = &btrfs_dir_file_operations; | |
39279cc3 | 4847 | |
dbe674a9 | 4848 | btrfs_i_size_write(inode, 0); |
39279cc3 CM |
4849 | err = btrfs_update_inode(trans, root, inode); |
4850 | if (err) | |
4851 | goto out_fail; | |
5f39d397 | 4852 | |
a1b075d2 JB |
4853 | err = btrfs_add_link(trans, dir, inode, dentry->d_name.name, |
4854 | dentry->d_name.len, 0, index); | |
39279cc3 CM |
4855 | if (err) |
4856 | goto out_fail; | |
5f39d397 | 4857 | |
39279cc3 CM |
4858 | d_instantiate(dentry, inode); |
4859 | drop_on_err = 0; | |
39279cc3 CM |
4860 | |
4861 | out_fail: | |
d3c2fdcf | 4862 | nr = trans->blocks_used; |
7ad85bb7 | 4863 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
4864 | if (drop_on_err) |
4865 | iput(inode); | |
d3c2fdcf | 4866 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
4867 | return err; |
4868 | } | |
4869 | ||
d352ac68 CM |
4870 | /* helper for btfs_get_extent. Given an existing extent in the tree, |
4871 | * and an extent that you want to insert, deal with overlap and insert | |
4872 | * the new extent into the tree. | |
4873 | */ | |
3b951516 CM |
4874 | static int merge_extent_mapping(struct extent_map_tree *em_tree, |
4875 | struct extent_map *existing, | |
e6dcd2dc CM |
4876 | struct extent_map *em, |
4877 | u64 map_start, u64 map_len) | |
3b951516 CM |
4878 | { |
4879 | u64 start_diff; | |
3b951516 | 4880 | |
e6dcd2dc CM |
4881 | BUG_ON(map_start < em->start || map_start >= extent_map_end(em)); |
4882 | start_diff = map_start - em->start; | |
4883 | em->start = map_start; | |
4884 | em->len = map_len; | |
c8b97818 CM |
4885 | if (em->block_start < EXTENT_MAP_LAST_BYTE && |
4886 | !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
e6dcd2dc | 4887 | em->block_start += start_diff; |
c8b97818 CM |
4888 | em->block_len -= start_diff; |
4889 | } | |
e6dcd2dc | 4890 | return add_extent_mapping(em_tree, em); |
3b951516 CM |
4891 | } |
4892 | ||
c8b97818 CM |
4893 | static noinline int uncompress_inline(struct btrfs_path *path, |
4894 | struct inode *inode, struct page *page, | |
4895 | size_t pg_offset, u64 extent_offset, | |
4896 | struct btrfs_file_extent_item *item) | |
4897 | { | |
4898 | int ret; | |
4899 | struct extent_buffer *leaf = path->nodes[0]; | |
4900 | char *tmp; | |
4901 | size_t max_size; | |
4902 | unsigned long inline_size; | |
4903 | unsigned long ptr; | |
261507a0 | 4904 | int compress_type; |
c8b97818 CM |
4905 | |
4906 | WARN_ON(pg_offset != 0); | |
261507a0 | 4907 | compress_type = btrfs_file_extent_compression(leaf, item); |
c8b97818 CM |
4908 | max_size = btrfs_file_extent_ram_bytes(leaf, item); |
4909 | inline_size = btrfs_file_extent_inline_item_len(leaf, | |
4910 | btrfs_item_nr(leaf, path->slots[0])); | |
4911 | tmp = kmalloc(inline_size, GFP_NOFS); | |
8d413713 TI |
4912 | if (!tmp) |
4913 | return -ENOMEM; | |
c8b97818 CM |
4914 | ptr = btrfs_file_extent_inline_start(item); |
4915 | ||
4916 | read_extent_buffer(leaf, tmp, ptr, inline_size); | |
4917 | ||
5b050f04 | 4918 | max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size); |
261507a0 LZ |
4919 | ret = btrfs_decompress(compress_type, tmp, page, |
4920 | extent_offset, inline_size, max_size); | |
c8b97818 CM |
4921 | if (ret) { |
4922 | char *kaddr = kmap_atomic(page, KM_USER0); | |
4923 | unsigned long copy_size = min_t(u64, | |
4924 | PAGE_CACHE_SIZE - pg_offset, | |
4925 | max_size - extent_offset); | |
4926 | memset(kaddr + pg_offset, 0, copy_size); | |
4927 | kunmap_atomic(kaddr, KM_USER0); | |
4928 | } | |
4929 | kfree(tmp); | |
4930 | return 0; | |
4931 | } | |
4932 | ||
d352ac68 CM |
4933 | /* |
4934 | * a bit scary, this does extent mapping from logical file offset to the disk. | |
d397712b CM |
4935 | * the ugly parts come from merging extents from the disk with the in-ram |
4936 | * representation. This gets more complex because of the data=ordered code, | |
d352ac68 CM |
4937 | * where the in-ram extents might be locked pending data=ordered completion. |
4938 | * | |
4939 | * This also copies inline extents directly into the page. | |
4940 | */ | |
d397712b | 4941 | |
a52d9a80 | 4942 | struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page, |
70dec807 | 4943 | size_t pg_offset, u64 start, u64 len, |
a52d9a80 CM |
4944 | int create) |
4945 | { | |
4946 | int ret; | |
4947 | int err = 0; | |
db94535d | 4948 | u64 bytenr; |
a52d9a80 CM |
4949 | u64 extent_start = 0; |
4950 | u64 extent_end = 0; | |
33345d01 | 4951 | u64 objectid = btrfs_ino(inode); |
a52d9a80 | 4952 | u32 found_type; |
f421950f | 4953 | struct btrfs_path *path = NULL; |
a52d9a80 CM |
4954 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4955 | struct btrfs_file_extent_item *item; | |
5f39d397 CM |
4956 | struct extent_buffer *leaf; |
4957 | struct btrfs_key found_key; | |
a52d9a80 CM |
4958 | struct extent_map *em = NULL; |
4959 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
d1310b2e | 4960 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
a52d9a80 | 4961 | struct btrfs_trans_handle *trans = NULL; |
261507a0 | 4962 | int compress_type; |
a52d9a80 | 4963 | |
a52d9a80 | 4964 | again: |
890871be | 4965 | read_lock(&em_tree->lock); |
d1310b2e | 4966 | em = lookup_extent_mapping(em_tree, start, len); |
a061fc8d CM |
4967 | if (em) |
4968 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
890871be | 4969 | read_unlock(&em_tree->lock); |
d1310b2e | 4970 | |
a52d9a80 | 4971 | if (em) { |
e1c4b745 CM |
4972 | if (em->start > start || em->start + em->len <= start) |
4973 | free_extent_map(em); | |
4974 | else if (em->block_start == EXTENT_MAP_INLINE && page) | |
70dec807 CM |
4975 | free_extent_map(em); |
4976 | else | |
4977 | goto out; | |
a52d9a80 | 4978 | } |
172ddd60 | 4979 | em = alloc_extent_map(); |
a52d9a80 | 4980 | if (!em) { |
d1310b2e CM |
4981 | err = -ENOMEM; |
4982 | goto out; | |
a52d9a80 | 4983 | } |
e6dcd2dc | 4984 | em->bdev = root->fs_info->fs_devices->latest_bdev; |
d1310b2e | 4985 | em->start = EXTENT_MAP_HOLE; |
445a6944 | 4986 | em->orig_start = EXTENT_MAP_HOLE; |
d1310b2e | 4987 | em->len = (u64)-1; |
c8b97818 | 4988 | em->block_len = (u64)-1; |
f421950f CM |
4989 | |
4990 | if (!path) { | |
4991 | path = btrfs_alloc_path(); | |
026fd317 JB |
4992 | if (!path) { |
4993 | err = -ENOMEM; | |
4994 | goto out; | |
4995 | } | |
4996 | /* | |
4997 | * Chances are we'll be called again, so go ahead and do | |
4998 | * readahead | |
4999 | */ | |
5000 | path->reada = 1; | |
f421950f CM |
5001 | } |
5002 | ||
179e29e4 CM |
5003 | ret = btrfs_lookup_file_extent(trans, root, path, |
5004 | objectid, start, trans != NULL); | |
a52d9a80 CM |
5005 | if (ret < 0) { |
5006 | err = ret; | |
5007 | goto out; | |
5008 | } | |
5009 | ||
5010 | if (ret != 0) { | |
5011 | if (path->slots[0] == 0) | |
5012 | goto not_found; | |
5013 | path->slots[0]--; | |
5014 | } | |
5015 | ||
5f39d397 CM |
5016 | leaf = path->nodes[0]; |
5017 | item = btrfs_item_ptr(leaf, path->slots[0], | |
a52d9a80 | 5018 | struct btrfs_file_extent_item); |
a52d9a80 | 5019 | /* are we inside the extent that was found? */ |
5f39d397 CM |
5020 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5021 | found_type = btrfs_key_type(&found_key); | |
5022 | if (found_key.objectid != objectid || | |
a52d9a80 CM |
5023 | found_type != BTRFS_EXTENT_DATA_KEY) { |
5024 | goto not_found; | |
5025 | } | |
5026 | ||
5f39d397 CM |
5027 | found_type = btrfs_file_extent_type(leaf, item); |
5028 | extent_start = found_key.offset; | |
261507a0 | 5029 | compress_type = btrfs_file_extent_compression(leaf, item); |
d899e052 YZ |
5030 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5031 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
a52d9a80 | 5032 | extent_end = extent_start + |
db94535d | 5033 | btrfs_file_extent_num_bytes(leaf, item); |
9036c102 YZ |
5034 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { |
5035 | size_t size; | |
5036 | size = btrfs_file_extent_inline_len(leaf, item); | |
5037 | extent_end = (extent_start + size + root->sectorsize - 1) & | |
5038 | ~((u64)root->sectorsize - 1); | |
5039 | } | |
5040 | ||
5041 | if (start >= extent_end) { | |
5042 | path->slots[0]++; | |
5043 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
5044 | ret = btrfs_next_leaf(root, path); | |
5045 | if (ret < 0) { | |
5046 | err = ret; | |
5047 | goto out; | |
a52d9a80 | 5048 | } |
9036c102 YZ |
5049 | if (ret > 0) |
5050 | goto not_found; | |
5051 | leaf = path->nodes[0]; | |
a52d9a80 | 5052 | } |
9036c102 YZ |
5053 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
5054 | if (found_key.objectid != objectid || | |
5055 | found_key.type != BTRFS_EXTENT_DATA_KEY) | |
5056 | goto not_found; | |
5057 | if (start + len <= found_key.offset) | |
5058 | goto not_found; | |
5059 | em->start = start; | |
5060 | em->len = found_key.offset - start; | |
5061 | goto not_found_em; | |
5062 | } | |
5063 | ||
d899e052 YZ |
5064 | if (found_type == BTRFS_FILE_EXTENT_REG || |
5065 | found_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
9036c102 YZ |
5066 | em->start = extent_start; |
5067 | em->len = extent_end - extent_start; | |
ff5b7ee3 YZ |
5068 | em->orig_start = extent_start - |
5069 | btrfs_file_extent_offset(leaf, item); | |
db94535d CM |
5070 | bytenr = btrfs_file_extent_disk_bytenr(leaf, item); |
5071 | if (bytenr == 0) { | |
5f39d397 | 5072 | em->block_start = EXTENT_MAP_HOLE; |
a52d9a80 CM |
5073 | goto insert; |
5074 | } | |
261507a0 | 5075 | if (compress_type != BTRFS_COMPRESS_NONE) { |
c8b97818 | 5076 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 | 5077 | em->compress_type = compress_type; |
c8b97818 CM |
5078 | em->block_start = bytenr; |
5079 | em->block_len = btrfs_file_extent_disk_num_bytes(leaf, | |
5080 | item); | |
5081 | } else { | |
5082 | bytenr += btrfs_file_extent_offset(leaf, item); | |
5083 | em->block_start = bytenr; | |
5084 | em->block_len = em->len; | |
d899e052 YZ |
5085 | if (found_type == BTRFS_FILE_EXTENT_PREALLOC) |
5086 | set_bit(EXTENT_FLAG_PREALLOC, &em->flags); | |
c8b97818 | 5087 | } |
a52d9a80 CM |
5088 | goto insert; |
5089 | } else if (found_type == BTRFS_FILE_EXTENT_INLINE) { | |
5f39d397 | 5090 | unsigned long ptr; |
a52d9a80 | 5091 | char *map; |
3326d1b0 CM |
5092 | size_t size; |
5093 | size_t extent_offset; | |
5094 | size_t copy_size; | |
a52d9a80 | 5095 | |
689f9346 | 5096 | em->block_start = EXTENT_MAP_INLINE; |
c8b97818 | 5097 | if (!page || create) { |
689f9346 | 5098 | em->start = extent_start; |
9036c102 | 5099 | em->len = extent_end - extent_start; |
689f9346 Y |
5100 | goto out; |
5101 | } | |
5f39d397 | 5102 | |
9036c102 YZ |
5103 | size = btrfs_file_extent_inline_len(leaf, item); |
5104 | extent_offset = page_offset(page) + pg_offset - extent_start; | |
70dec807 | 5105 | copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset, |
3326d1b0 | 5106 | size - extent_offset); |
3326d1b0 | 5107 | em->start = extent_start + extent_offset; |
70dec807 CM |
5108 | em->len = (copy_size + root->sectorsize - 1) & |
5109 | ~((u64)root->sectorsize - 1); | |
ff5b7ee3 | 5110 | em->orig_start = EXTENT_MAP_INLINE; |
261507a0 | 5111 | if (compress_type) { |
c8b97818 | 5112 | set_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
261507a0 LZ |
5113 | em->compress_type = compress_type; |
5114 | } | |
689f9346 | 5115 | ptr = btrfs_file_extent_inline_start(item) + extent_offset; |
179e29e4 | 5116 | if (create == 0 && !PageUptodate(page)) { |
261507a0 LZ |
5117 | if (btrfs_file_extent_compression(leaf, item) != |
5118 | BTRFS_COMPRESS_NONE) { | |
c8b97818 CM |
5119 | ret = uncompress_inline(path, inode, page, |
5120 | pg_offset, | |
5121 | extent_offset, item); | |
5122 | BUG_ON(ret); | |
5123 | } else { | |
5124 | map = kmap(page); | |
5125 | read_extent_buffer(leaf, map + pg_offset, ptr, | |
5126 | copy_size); | |
93c82d57 CM |
5127 | if (pg_offset + copy_size < PAGE_CACHE_SIZE) { |
5128 | memset(map + pg_offset + copy_size, 0, | |
5129 | PAGE_CACHE_SIZE - pg_offset - | |
5130 | copy_size); | |
5131 | } | |
c8b97818 CM |
5132 | kunmap(page); |
5133 | } | |
179e29e4 CM |
5134 | flush_dcache_page(page); |
5135 | } else if (create && PageUptodate(page)) { | |
6bf7e080 | 5136 | BUG(); |
179e29e4 CM |
5137 | if (!trans) { |
5138 | kunmap(page); | |
5139 | free_extent_map(em); | |
5140 | em = NULL; | |
ff5714cc | 5141 | |
b3b4aa74 | 5142 | btrfs_release_path(path); |
7a7eaa40 | 5143 | trans = btrfs_join_transaction(root); |
ff5714cc | 5144 | |
3612b495 TI |
5145 | if (IS_ERR(trans)) |
5146 | return ERR_CAST(trans); | |
179e29e4 CM |
5147 | goto again; |
5148 | } | |
c8b97818 | 5149 | map = kmap(page); |
70dec807 | 5150 | write_extent_buffer(leaf, map + pg_offset, ptr, |
179e29e4 | 5151 | copy_size); |
c8b97818 | 5152 | kunmap(page); |
179e29e4 | 5153 | btrfs_mark_buffer_dirty(leaf); |
a52d9a80 | 5154 | } |
d1310b2e | 5155 | set_extent_uptodate(io_tree, em->start, |
507903b8 | 5156 | extent_map_end(em) - 1, NULL, GFP_NOFS); |
a52d9a80 CM |
5157 | goto insert; |
5158 | } else { | |
d397712b | 5159 | printk(KERN_ERR "btrfs unknown found_type %d\n", found_type); |
a52d9a80 CM |
5160 | WARN_ON(1); |
5161 | } | |
5162 | not_found: | |
5163 | em->start = start; | |
d1310b2e | 5164 | em->len = len; |
a52d9a80 | 5165 | not_found_em: |
5f39d397 | 5166 | em->block_start = EXTENT_MAP_HOLE; |
9036c102 | 5167 | set_bit(EXTENT_FLAG_VACANCY, &em->flags); |
a52d9a80 | 5168 | insert: |
b3b4aa74 | 5169 | btrfs_release_path(path); |
d1310b2e | 5170 | if (em->start > start || extent_map_end(em) <= start) { |
d397712b CM |
5171 | printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed " |
5172 | "[%llu %llu]\n", (unsigned long long)em->start, | |
5173 | (unsigned long long)em->len, | |
5174 | (unsigned long long)start, | |
5175 | (unsigned long long)len); | |
a52d9a80 CM |
5176 | err = -EIO; |
5177 | goto out; | |
5178 | } | |
d1310b2e CM |
5179 | |
5180 | err = 0; | |
890871be | 5181 | write_lock(&em_tree->lock); |
a52d9a80 | 5182 | ret = add_extent_mapping(em_tree, em); |
3b951516 CM |
5183 | /* it is possible that someone inserted the extent into the tree |
5184 | * while we had the lock dropped. It is also possible that | |
5185 | * an overlapping map exists in the tree | |
5186 | */ | |
a52d9a80 | 5187 | if (ret == -EEXIST) { |
3b951516 | 5188 | struct extent_map *existing; |
e6dcd2dc CM |
5189 | |
5190 | ret = 0; | |
5191 | ||
3b951516 | 5192 | existing = lookup_extent_mapping(em_tree, start, len); |
e1c4b745 CM |
5193 | if (existing && (existing->start > start || |
5194 | existing->start + existing->len <= start)) { | |
5195 | free_extent_map(existing); | |
5196 | existing = NULL; | |
5197 | } | |
3b951516 CM |
5198 | if (!existing) { |
5199 | existing = lookup_extent_mapping(em_tree, em->start, | |
5200 | em->len); | |
5201 | if (existing) { | |
5202 | err = merge_extent_mapping(em_tree, existing, | |
e6dcd2dc CM |
5203 | em, start, |
5204 | root->sectorsize); | |
3b951516 CM |
5205 | free_extent_map(existing); |
5206 | if (err) { | |
5207 | free_extent_map(em); | |
5208 | em = NULL; | |
5209 | } | |
5210 | } else { | |
5211 | err = -EIO; | |
3b951516 CM |
5212 | free_extent_map(em); |
5213 | em = NULL; | |
5214 | } | |
5215 | } else { | |
5216 | free_extent_map(em); | |
5217 | em = existing; | |
e6dcd2dc | 5218 | err = 0; |
a52d9a80 | 5219 | } |
a52d9a80 | 5220 | } |
890871be | 5221 | write_unlock(&em_tree->lock); |
a52d9a80 | 5222 | out: |
1abe9b8a | 5223 | |
5224 | trace_btrfs_get_extent(root, em); | |
5225 | ||
f421950f CM |
5226 | if (path) |
5227 | btrfs_free_path(path); | |
a52d9a80 CM |
5228 | if (trans) { |
5229 | ret = btrfs_end_transaction(trans, root); | |
d397712b | 5230 | if (!err) |
a52d9a80 CM |
5231 | err = ret; |
5232 | } | |
a52d9a80 CM |
5233 | if (err) { |
5234 | free_extent_map(em); | |
a52d9a80 CM |
5235 | return ERR_PTR(err); |
5236 | } | |
5237 | return em; | |
5238 | } | |
5239 | ||
ec29ed5b CM |
5240 | struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page, |
5241 | size_t pg_offset, u64 start, u64 len, | |
5242 | int create) | |
5243 | { | |
5244 | struct extent_map *em; | |
5245 | struct extent_map *hole_em = NULL; | |
5246 | u64 range_start = start; | |
5247 | u64 end; | |
5248 | u64 found; | |
5249 | u64 found_end; | |
5250 | int err = 0; | |
5251 | ||
5252 | em = btrfs_get_extent(inode, page, pg_offset, start, len, create); | |
5253 | if (IS_ERR(em)) | |
5254 | return em; | |
5255 | if (em) { | |
5256 | /* | |
5257 | * if our em maps to a hole, there might | |
5258 | * actually be delalloc bytes behind it | |
5259 | */ | |
5260 | if (em->block_start != EXTENT_MAP_HOLE) | |
5261 | return em; | |
5262 | else | |
5263 | hole_em = em; | |
5264 | } | |
5265 | ||
5266 | /* check to see if we've wrapped (len == -1 or similar) */ | |
5267 | end = start + len; | |
5268 | if (end < start) | |
5269 | end = (u64)-1; | |
5270 | else | |
5271 | end -= 1; | |
5272 | ||
5273 | em = NULL; | |
5274 | ||
5275 | /* ok, we didn't find anything, lets look for delalloc */ | |
5276 | found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start, | |
5277 | end, len, EXTENT_DELALLOC, 1); | |
5278 | found_end = range_start + found; | |
5279 | if (found_end < range_start) | |
5280 | found_end = (u64)-1; | |
5281 | ||
5282 | /* | |
5283 | * we didn't find anything useful, return | |
5284 | * the original results from get_extent() | |
5285 | */ | |
5286 | if (range_start > end || found_end <= start) { | |
5287 | em = hole_em; | |
5288 | hole_em = NULL; | |
5289 | goto out; | |
5290 | } | |
5291 | ||
5292 | /* adjust the range_start to make sure it doesn't | |
5293 | * go backwards from the start they passed in | |
5294 | */ | |
5295 | range_start = max(start,range_start); | |
5296 | found = found_end - range_start; | |
5297 | ||
5298 | if (found > 0) { | |
5299 | u64 hole_start = start; | |
5300 | u64 hole_len = len; | |
5301 | ||
172ddd60 | 5302 | em = alloc_extent_map(); |
ec29ed5b CM |
5303 | if (!em) { |
5304 | err = -ENOMEM; | |
5305 | goto out; | |
5306 | } | |
5307 | /* | |
5308 | * when btrfs_get_extent can't find anything it | |
5309 | * returns one huge hole | |
5310 | * | |
5311 | * make sure what it found really fits our range, and | |
5312 | * adjust to make sure it is based on the start from | |
5313 | * the caller | |
5314 | */ | |
5315 | if (hole_em) { | |
5316 | u64 calc_end = extent_map_end(hole_em); | |
5317 | ||
5318 | if (calc_end <= start || (hole_em->start > end)) { | |
5319 | free_extent_map(hole_em); | |
5320 | hole_em = NULL; | |
5321 | } else { | |
5322 | hole_start = max(hole_em->start, start); | |
5323 | hole_len = calc_end - hole_start; | |
5324 | } | |
5325 | } | |
5326 | em->bdev = NULL; | |
5327 | if (hole_em && range_start > hole_start) { | |
5328 | /* our hole starts before our delalloc, so we | |
5329 | * have to return just the parts of the hole | |
5330 | * that go until the delalloc starts | |
5331 | */ | |
5332 | em->len = min(hole_len, | |
5333 | range_start - hole_start); | |
5334 | em->start = hole_start; | |
5335 | em->orig_start = hole_start; | |
5336 | /* | |
5337 | * don't adjust block start at all, | |
5338 | * it is fixed at EXTENT_MAP_HOLE | |
5339 | */ | |
5340 | em->block_start = hole_em->block_start; | |
5341 | em->block_len = hole_len; | |
5342 | } else { | |
5343 | em->start = range_start; | |
5344 | em->len = found; | |
5345 | em->orig_start = range_start; | |
5346 | em->block_start = EXTENT_MAP_DELALLOC; | |
5347 | em->block_len = found; | |
5348 | } | |
5349 | } else if (hole_em) { | |
5350 | return hole_em; | |
5351 | } | |
5352 | out: | |
5353 | ||
5354 | free_extent_map(hole_em); | |
5355 | if (err) { | |
5356 | free_extent_map(em); | |
5357 | return ERR_PTR(err); | |
5358 | } | |
5359 | return em; | |
5360 | } | |
5361 | ||
4b46fce2 | 5362 | static struct extent_map *btrfs_new_extent_direct(struct inode *inode, |
16d299ac | 5363 | struct extent_map *em, |
4b46fce2 JB |
5364 | u64 start, u64 len) |
5365 | { | |
5366 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5367 | struct btrfs_trans_handle *trans; | |
4b46fce2 JB |
5368 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
5369 | struct btrfs_key ins; | |
5370 | u64 alloc_hint; | |
5371 | int ret; | |
16d299ac | 5372 | bool insert = false; |
4b46fce2 | 5373 | |
16d299ac JB |
5374 | /* |
5375 | * Ok if the extent map we looked up is a hole and is for the exact | |
5376 | * range we want, there is no reason to allocate a new one, however if | |
5377 | * it is not right then we need to free this one and drop the cache for | |
5378 | * our range. | |
5379 | */ | |
5380 | if (em->block_start != EXTENT_MAP_HOLE || em->start != start || | |
5381 | em->len != len) { | |
5382 | free_extent_map(em); | |
5383 | em = NULL; | |
5384 | insert = true; | |
5385 | btrfs_drop_extent_cache(inode, start, start + len - 1, 0); | |
5386 | } | |
4b46fce2 | 5387 | |
7a7eaa40 | 5388 | trans = btrfs_join_transaction(root); |
3612b495 TI |
5389 | if (IS_ERR(trans)) |
5390 | return ERR_CAST(trans); | |
4b46fce2 | 5391 | |
4cb5300b CM |
5392 | if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024) |
5393 | btrfs_add_inode_defrag(trans, inode); | |
5394 | ||
4b46fce2 JB |
5395 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; |
5396 | ||
5397 | alloc_hint = get_extent_allocation_hint(inode, start, len); | |
5398 | ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0, | |
5399 | alloc_hint, (u64)-1, &ins, 1); | |
5400 | if (ret) { | |
5401 | em = ERR_PTR(ret); | |
5402 | goto out; | |
5403 | } | |
5404 | ||
4b46fce2 | 5405 | if (!em) { |
172ddd60 | 5406 | em = alloc_extent_map(); |
16d299ac JB |
5407 | if (!em) { |
5408 | em = ERR_PTR(-ENOMEM); | |
5409 | goto out; | |
5410 | } | |
4b46fce2 JB |
5411 | } |
5412 | ||
5413 | em->start = start; | |
5414 | em->orig_start = em->start; | |
5415 | em->len = ins.offset; | |
5416 | ||
5417 | em->block_start = ins.objectid; | |
5418 | em->block_len = ins.offset; | |
5419 | em->bdev = root->fs_info->fs_devices->latest_bdev; | |
16d299ac JB |
5420 | |
5421 | /* | |
5422 | * We need to do this because if we're using the original em we searched | |
5423 | * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that. | |
5424 | */ | |
5425 | em->flags = 0; | |
4b46fce2 JB |
5426 | set_bit(EXTENT_FLAG_PINNED, &em->flags); |
5427 | ||
16d299ac | 5428 | while (insert) { |
4b46fce2 JB |
5429 | write_lock(&em_tree->lock); |
5430 | ret = add_extent_mapping(em_tree, em); | |
5431 | write_unlock(&em_tree->lock); | |
5432 | if (ret != -EEXIST) | |
5433 | break; | |
5434 | btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0); | |
5435 | } | |
5436 | ||
5437 | ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid, | |
5438 | ins.offset, ins.offset, 0); | |
5439 | if (ret) { | |
5440 | btrfs_free_reserved_extent(root, ins.objectid, ins.offset); | |
5441 | em = ERR_PTR(ret); | |
5442 | } | |
5443 | out: | |
5444 | btrfs_end_transaction(trans, root); | |
5445 | return em; | |
5446 | } | |
5447 | ||
46bfbb5c CM |
5448 | /* |
5449 | * returns 1 when the nocow is safe, < 1 on error, 0 if the | |
5450 | * block must be cow'd | |
5451 | */ | |
5452 | static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans, | |
5453 | struct inode *inode, u64 offset, u64 len) | |
5454 | { | |
5455 | struct btrfs_path *path; | |
5456 | int ret; | |
5457 | struct extent_buffer *leaf; | |
5458 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5459 | struct btrfs_file_extent_item *fi; | |
5460 | struct btrfs_key key; | |
5461 | u64 disk_bytenr; | |
5462 | u64 backref_offset; | |
5463 | u64 extent_end; | |
5464 | u64 num_bytes; | |
5465 | int slot; | |
5466 | int found_type; | |
5467 | ||
5468 | path = btrfs_alloc_path(); | |
5469 | if (!path) | |
5470 | return -ENOMEM; | |
5471 | ||
33345d01 | 5472 | ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode), |
46bfbb5c CM |
5473 | offset, 0); |
5474 | if (ret < 0) | |
5475 | goto out; | |
5476 | ||
5477 | slot = path->slots[0]; | |
5478 | if (ret == 1) { | |
5479 | if (slot == 0) { | |
5480 | /* can't find the item, must cow */ | |
5481 | ret = 0; | |
5482 | goto out; | |
5483 | } | |
5484 | slot--; | |
5485 | } | |
5486 | ret = 0; | |
5487 | leaf = path->nodes[0]; | |
5488 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
33345d01 | 5489 | if (key.objectid != btrfs_ino(inode) || |
46bfbb5c CM |
5490 | key.type != BTRFS_EXTENT_DATA_KEY) { |
5491 | /* not our file or wrong item type, must cow */ | |
5492 | goto out; | |
5493 | } | |
5494 | ||
5495 | if (key.offset > offset) { | |
5496 | /* Wrong offset, must cow */ | |
5497 | goto out; | |
5498 | } | |
5499 | ||
5500 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); | |
5501 | found_type = btrfs_file_extent_type(leaf, fi); | |
5502 | if (found_type != BTRFS_FILE_EXTENT_REG && | |
5503 | found_type != BTRFS_FILE_EXTENT_PREALLOC) { | |
5504 | /* not a regular extent, must cow */ | |
5505 | goto out; | |
5506 | } | |
5507 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); | |
5508 | backref_offset = btrfs_file_extent_offset(leaf, fi); | |
5509 | ||
5510 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); | |
5511 | if (extent_end < offset + len) { | |
5512 | /* extent doesn't include our full range, must cow */ | |
5513 | goto out; | |
5514 | } | |
5515 | ||
5516 | if (btrfs_extent_readonly(root, disk_bytenr)) | |
5517 | goto out; | |
5518 | ||
5519 | /* | |
5520 | * look for other files referencing this extent, if we | |
5521 | * find any we must cow | |
5522 | */ | |
33345d01 | 5523 | if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode), |
46bfbb5c CM |
5524 | key.offset - backref_offset, disk_bytenr)) |
5525 | goto out; | |
5526 | ||
5527 | /* | |
5528 | * adjust disk_bytenr and num_bytes to cover just the bytes | |
5529 | * in this extent we are about to write. If there | |
5530 | * are any csums in that range we have to cow in order | |
5531 | * to keep the csums correct | |
5532 | */ | |
5533 | disk_bytenr += backref_offset; | |
5534 | disk_bytenr += offset - key.offset; | |
5535 | num_bytes = min(offset + len, extent_end) - offset; | |
5536 | if (csum_exist_in_range(root, disk_bytenr, num_bytes)) | |
5537 | goto out; | |
5538 | /* | |
5539 | * all of the above have passed, it is safe to overwrite this extent | |
5540 | * without cow | |
5541 | */ | |
5542 | ret = 1; | |
5543 | out: | |
5544 | btrfs_free_path(path); | |
5545 | return ret; | |
5546 | } | |
5547 | ||
4b46fce2 JB |
5548 | static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock, |
5549 | struct buffer_head *bh_result, int create) | |
5550 | { | |
5551 | struct extent_map *em; | |
5552 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5553 | u64 start = iblock << inode->i_blkbits; | |
5554 | u64 len = bh_result->b_size; | |
46bfbb5c | 5555 | struct btrfs_trans_handle *trans; |
4b46fce2 JB |
5556 | |
5557 | em = btrfs_get_extent(inode, NULL, 0, start, len, 0); | |
5558 | if (IS_ERR(em)) | |
5559 | return PTR_ERR(em); | |
5560 | ||
5561 | /* | |
5562 | * Ok for INLINE and COMPRESSED extents we need to fallback on buffered | |
5563 | * io. INLINE is special, and we could probably kludge it in here, but | |
5564 | * it's still buffered so for safety lets just fall back to the generic | |
5565 | * buffered path. | |
5566 | * | |
5567 | * For COMPRESSED we _have_ to read the entire extent in so we can | |
5568 | * decompress it, so there will be buffering required no matter what we | |
5569 | * do, so go ahead and fallback to buffered. | |
5570 | * | |
5571 | * We return -ENOTBLK because thats what makes DIO go ahead and go back | |
5572 | * to buffered IO. Don't blame me, this is the price we pay for using | |
5573 | * the generic code. | |
5574 | */ | |
5575 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) || | |
5576 | em->block_start == EXTENT_MAP_INLINE) { | |
5577 | free_extent_map(em); | |
5578 | return -ENOTBLK; | |
5579 | } | |
5580 | ||
5581 | /* Just a good old fashioned hole, return */ | |
5582 | if (!create && (em->block_start == EXTENT_MAP_HOLE || | |
5583 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { | |
5584 | free_extent_map(em); | |
5585 | /* DIO will do one hole at a time, so just unlock a sector */ | |
5586 | unlock_extent(&BTRFS_I(inode)->io_tree, start, | |
5587 | start + root->sectorsize - 1, GFP_NOFS); | |
5588 | return 0; | |
5589 | } | |
5590 | ||
5591 | /* | |
5592 | * We don't allocate a new extent in the following cases | |
5593 | * | |
5594 | * 1) The inode is marked as NODATACOW. In this case we'll just use the | |
5595 | * existing extent. | |
5596 | * 2) The extent is marked as PREALLOC. We're good to go here and can | |
5597 | * just use the extent. | |
5598 | * | |
5599 | */ | |
46bfbb5c CM |
5600 | if (!create) { |
5601 | len = em->len - (start - em->start); | |
4b46fce2 | 5602 | goto map; |
46bfbb5c | 5603 | } |
4b46fce2 JB |
5604 | |
5605 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) || | |
5606 | ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) && | |
5607 | em->block_start != EXTENT_MAP_HOLE)) { | |
4b46fce2 JB |
5608 | int type; |
5609 | int ret; | |
46bfbb5c | 5610 | u64 block_start; |
4b46fce2 JB |
5611 | |
5612 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5613 | type = BTRFS_ORDERED_PREALLOC; | |
5614 | else | |
5615 | type = BTRFS_ORDERED_NOCOW; | |
46bfbb5c | 5616 | len = min(len, em->len - (start - em->start)); |
4b46fce2 | 5617 | block_start = em->block_start + (start - em->start); |
46bfbb5c CM |
5618 | |
5619 | /* | |
5620 | * we're not going to log anything, but we do need | |
5621 | * to make sure the current transaction stays open | |
5622 | * while we look for nocow cross refs | |
5623 | */ | |
7a7eaa40 | 5624 | trans = btrfs_join_transaction(root); |
3612b495 | 5625 | if (IS_ERR(trans)) |
46bfbb5c CM |
5626 | goto must_cow; |
5627 | ||
5628 | if (can_nocow_odirect(trans, inode, start, len) == 1) { | |
5629 | ret = btrfs_add_ordered_extent_dio(inode, start, | |
5630 | block_start, len, len, type); | |
5631 | btrfs_end_transaction(trans, root); | |
5632 | if (ret) { | |
5633 | free_extent_map(em); | |
5634 | return ret; | |
5635 | } | |
5636 | goto unlock; | |
4b46fce2 | 5637 | } |
46bfbb5c | 5638 | btrfs_end_transaction(trans, root); |
4b46fce2 | 5639 | } |
46bfbb5c CM |
5640 | must_cow: |
5641 | /* | |
5642 | * this will cow the extent, reset the len in case we changed | |
5643 | * it above | |
5644 | */ | |
5645 | len = bh_result->b_size; | |
16d299ac | 5646 | em = btrfs_new_extent_direct(inode, em, start, len); |
46bfbb5c CM |
5647 | if (IS_ERR(em)) |
5648 | return PTR_ERR(em); | |
5649 | len = min(len, em->len - (start - em->start)); | |
5650 | unlock: | |
4845e44f CM |
5651 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1, |
5652 | EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1, | |
5653 | 0, NULL, GFP_NOFS); | |
4b46fce2 JB |
5654 | map: |
5655 | bh_result->b_blocknr = (em->block_start + (start - em->start)) >> | |
5656 | inode->i_blkbits; | |
46bfbb5c | 5657 | bh_result->b_size = len; |
4b46fce2 JB |
5658 | bh_result->b_bdev = em->bdev; |
5659 | set_buffer_mapped(bh_result); | |
5660 | if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) | |
5661 | set_buffer_new(bh_result); | |
5662 | ||
5663 | free_extent_map(em); | |
5664 | ||
5665 | return 0; | |
5666 | } | |
5667 | ||
5668 | struct btrfs_dio_private { | |
5669 | struct inode *inode; | |
5670 | u64 logical_offset; | |
5671 | u64 disk_bytenr; | |
5672 | u64 bytes; | |
5673 | u32 *csums; | |
5674 | void *private; | |
e65e1535 MX |
5675 | |
5676 | /* number of bios pending for this dio */ | |
5677 | atomic_t pending_bios; | |
5678 | ||
5679 | /* IO errors */ | |
5680 | int errors; | |
5681 | ||
5682 | struct bio *orig_bio; | |
4b46fce2 JB |
5683 | }; |
5684 | ||
5685 | static void btrfs_endio_direct_read(struct bio *bio, int err) | |
5686 | { | |
e65e1535 | 5687 | struct btrfs_dio_private *dip = bio->bi_private; |
4b46fce2 JB |
5688 | struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1; |
5689 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 JB |
5690 | struct inode *inode = dip->inode; |
5691 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5692 | u64 start; | |
5693 | u32 *private = dip->csums; | |
5694 | ||
5695 | start = dip->logical_offset; | |
5696 | do { | |
5697 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { | |
5698 | struct page *page = bvec->bv_page; | |
5699 | char *kaddr; | |
5700 | u32 csum = ~(u32)0; | |
5701 | unsigned long flags; | |
5702 | ||
5703 | local_irq_save(flags); | |
5704 | kaddr = kmap_atomic(page, KM_IRQ0); | |
5705 | csum = btrfs_csum_data(root, kaddr + bvec->bv_offset, | |
5706 | csum, bvec->bv_len); | |
5707 | btrfs_csum_final(csum, (char *)&csum); | |
5708 | kunmap_atomic(kaddr, KM_IRQ0); | |
5709 | local_irq_restore(flags); | |
5710 | ||
5711 | flush_dcache_page(bvec->bv_page); | |
5712 | if (csum != *private) { | |
33345d01 | 5713 | printk(KERN_ERR "btrfs csum failed ino %llu off" |
4b46fce2 | 5714 | " %llu csum %u private %u\n", |
33345d01 LZ |
5715 | (unsigned long long)btrfs_ino(inode), |
5716 | (unsigned long long)start, | |
4b46fce2 JB |
5717 | csum, *private); |
5718 | err = -EIO; | |
5719 | } | |
5720 | } | |
5721 | ||
5722 | start += bvec->bv_len; | |
5723 | private++; | |
5724 | bvec++; | |
5725 | } while (bvec <= bvec_end); | |
5726 | ||
5727 | unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset, | |
5728 | dip->logical_offset + dip->bytes - 1, GFP_NOFS); | |
5729 | bio->bi_private = dip->private; | |
5730 | ||
5731 | kfree(dip->csums); | |
5732 | kfree(dip); | |
c0da7aa1 JB |
5733 | |
5734 | /* If we had a csum failure make sure to clear the uptodate flag */ | |
5735 | if (err) | |
5736 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5737 | dio_end_io(bio, err); |
5738 | } | |
5739 | ||
5740 | static void btrfs_endio_direct_write(struct bio *bio, int err) | |
5741 | { | |
5742 | struct btrfs_dio_private *dip = bio->bi_private; | |
5743 | struct inode *inode = dip->inode; | |
5744 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5745 | struct btrfs_trans_handle *trans; | |
5746 | struct btrfs_ordered_extent *ordered = NULL; | |
5747 | struct extent_state *cached_state = NULL; | |
163cf09c CM |
5748 | u64 ordered_offset = dip->logical_offset; |
5749 | u64 ordered_bytes = dip->bytes; | |
4b46fce2 JB |
5750 | int ret; |
5751 | ||
5752 | if (err) | |
5753 | goto out_done; | |
163cf09c CM |
5754 | again: |
5755 | ret = btrfs_dec_test_first_ordered_pending(inode, &ordered, | |
5756 | &ordered_offset, | |
5757 | ordered_bytes); | |
4b46fce2 | 5758 | if (!ret) |
163cf09c | 5759 | goto out_test; |
4b46fce2 JB |
5760 | |
5761 | BUG_ON(!ordered); | |
5762 | ||
7a7eaa40 | 5763 | trans = btrfs_join_transaction(root); |
3612b495 | 5764 | if (IS_ERR(trans)) { |
4b46fce2 JB |
5765 | err = -ENOMEM; |
5766 | goto out; | |
5767 | } | |
5768 | trans->block_rsv = &root->fs_info->delalloc_block_rsv; | |
5769 | ||
5770 | if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) { | |
5771 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); | |
5772 | if (!ret) | |
2115133f | 5773 | err = btrfs_update_inode_fallback(trans, root, inode); |
4b46fce2 JB |
5774 | goto out; |
5775 | } | |
5776 | ||
5777 | lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5778 | ordered->file_offset + ordered->len - 1, 0, | |
5779 | &cached_state, GFP_NOFS); | |
5780 | ||
5781 | if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) { | |
5782 | ret = btrfs_mark_extent_written(trans, inode, | |
5783 | ordered->file_offset, | |
5784 | ordered->file_offset + | |
5785 | ordered->len); | |
5786 | if (ret) { | |
5787 | err = ret; | |
5788 | goto out_unlock; | |
5789 | } | |
5790 | } else { | |
5791 | ret = insert_reserved_file_extent(trans, inode, | |
5792 | ordered->file_offset, | |
5793 | ordered->start, | |
5794 | ordered->disk_len, | |
5795 | ordered->len, | |
5796 | ordered->len, | |
5797 | 0, 0, 0, | |
5798 | BTRFS_FILE_EXTENT_REG); | |
5799 | unpin_extent_cache(&BTRFS_I(inode)->extent_tree, | |
5800 | ordered->file_offset, ordered->len); | |
5801 | if (ret) { | |
5802 | err = ret; | |
5803 | WARN_ON(1); | |
5804 | goto out_unlock; | |
5805 | } | |
5806 | } | |
5807 | ||
5808 | add_pending_csums(trans, inode, ordered->file_offset, &ordered->list); | |
1ef30be1 | 5809 | ret = btrfs_ordered_update_i_size(inode, 0, ordered); |
a39f7521 | 5810 | if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) |
2115133f | 5811 | btrfs_update_inode_fallback(trans, root, inode); |
1ef30be1 | 5812 | ret = 0; |
4b46fce2 JB |
5813 | out_unlock: |
5814 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset, | |
5815 | ordered->file_offset + ordered->len - 1, | |
5816 | &cached_state, GFP_NOFS); | |
5817 | out: | |
5818 | btrfs_delalloc_release_metadata(inode, ordered->len); | |
5819 | btrfs_end_transaction(trans, root); | |
163cf09c | 5820 | ordered_offset = ordered->file_offset + ordered->len; |
4b46fce2 JB |
5821 | btrfs_put_ordered_extent(ordered); |
5822 | btrfs_put_ordered_extent(ordered); | |
163cf09c CM |
5823 | |
5824 | out_test: | |
5825 | /* | |
5826 | * our bio might span multiple ordered extents. If we haven't | |
5827 | * completed the accounting for the whole dio, go back and try again | |
5828 | */ | |
5829 | if (ordered_offset < dip->logical_offset + dip->bytes) { | |
5830 | ordered_bytes = dip->logical_offset + dip->bytes - | |
5831 | ordered_offset; | |
5832 | goto again; | |
5833 | } | |
4b46fce2 JB |
5834 | out_done: |
5835 | bio->bi_private = dip->private; | |
5836 | ||
5837 | kfree(dip->csums); | |
5838 | kfree(dip); | |
c0da7aa1 JB |
5839 | |
5840 | /* If we had an error make sure to clear the uptodate flag */ | |
5841 | if (err) | |
5842 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
4b46fce2 JB |
5843 | dio_end_io(bio, err); |
5844 | } | |
5845 | ||
eaf25d93 CM |
5846 | static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw, |
5847 | struct bio *bio, int mirror_num, | |
5848 | unsigned long bio_flags, u64 offset) | |
5849 | { | |
5850 | int ret; | |
5851 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5852 | ret = btrfs_csum_one_bio(root, inode, bio, offset, 1); | |
5853 | BUG_ON(ret); | |
5854 | return 0; | |
5855 | } | |
5856 | ||
e65e1535 MX |
5857 | static void btrfs_end_dio_bio(struct bio *bio, int err) |
5858 | { | |
5859 | struct btrfs_dio_private *dip = bio->bi_private; | |
5860 | ||
5861 | if (err) { | |
33345d01 | 5862 | printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu " |
3dd1462e | 5863 | "sector %#Lx len %u err no %d\n", |
33345d01 | 5864 | (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw, |
3dd1462e | 5865 | (unsigned long long)bio->bi_sector, bio->bi_size, err); |
e65e1535 MX |
5866 | dip->errors = 1; |
5867 | ||
5868 | /* | |
5869 | * before atomic variable goto zero, we must make sure | |
5870 | * dip->errors is perceived to be set. | |
5871 | */ | |
5872 | smp_mb__before_atomic_dec(); | |
5873 | } | |
5874 | ||
5875 | /* if there are more bios still pending for this dio, just exit */ | |
5876 | if (!atomic_dec_and_test(&dip->pending_bios)) | |
5877 | goto out; | |
5878 | ||
5879 | if (dip->errors) | |
5880 | bio_io_error(dip->orig_bio); | |
5881 | else { | |
5882 | set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags); | |
5883 | bio_endio(dip->orig_bio, 0); | |
5884 | } | |
5885 | out: | |
5886 | bio_put(bio); | |
5887 | } | |
5888 | ||
5889 | static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev, | |
5890 | u64 first_sector, gfp_t gfp_flags) | |
5891 | { | |
5892 | int nr_vecs = bio_get_nr_vecs(bdev); | |
5893 | return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags); | |
5894 | } | |
5895 | ||
5896 | static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, | |
5897 | int rw, u64 file_offset, int skip_sum, | |
1ae39938 | 5898 | u32 *csums, int async_submit) |
e65e1535 MX |
5899 | { |
5900 | int write = rw & REQ_WRITE; | |
5901 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5902 | int ret; | |
5903 | ||
5904 | bio_get(bio); | |
5905 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
5906 | if (ret) | |
5907 | goto err; | |
5908 | ||
1ae39938 JB |
5909 | if (skip_sum) |
5910 | goto map; | |
5911 | ||
5912 | if (write && async_submit) { | |
e65e1535 MX |
5913 | ret = btrfs_wq_submit_bio(root->fs_info, |
5914 | inode, rw, bio, 0, 0, | |
5915 | file_offset, | |
5916 | __btrfs_submit_bio_start_direct_io, | |
5917 | __btrfs_submit_bio_done); | |
5918 | goto err; | |
1ae39938 JB |
5919 | } else if (write) { |
5920 | /* | |
5921 | * If we aren't doing async submit, calculate the csum of the | |
5922 | * bio now. | |
5923 | */ | |
5924 | ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1); | |
5925 | if (ret) | |
5926 | goto err; | |
c2db1073 TI |
5927 | } else if (!skip_sum) { |
5928 | ret = btrfs_lookup_bio_sums_dio(root, inode, bio, | |
e65e1535 | 5929 | file_offset, csums); |
c2db1073 TI |
5930 | if (ret) |
5931 | goto err; | |
5932 | } | |
e65e1535 | 5933 | |
1ae39938 JB |
5934 | map: |
5935 | ret = btrfs_map_bio(root, rw, bio, 0, async_submit); | |
e65e1535 MX |
5936 | err: |
5937 | bio_put(bio); | |
5938 | return ret; | |
5939 | } | |
5940 | ||
5941 | static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip, | |
5942 | int skip_sum) | |
5943 | { | |
5944 | struct inode *inode = dip->inode; | |
5945 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
5946 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
5947 | struct bio *bio; | |
5948 | struct bio *orig_bio = dip->orig_bio; | |
5949 | struct bio_vec *bvec = orig_bio->bi_io_vec; | |
5950 | u64 start_sector = orig_bio->bi_sector; | |
5951 | u64 file_offset = dip->logical_offset; | |
5952 | u64 submit_len = 0; | |
5953 | u64 map_length; | |
5954 | int nr_pages = 0; | |
5955 | u32 *csums = dip->csums; | |
5956 | int ret = 0; | |
1ae39938 | 5957 | int async_submit = 0; |
98bc3149 | 5958 | int write = rw & REQ_WRITE; |
e65e1535 | 5959 | |
e65e1535 MX |
5960 | map_length = orig_bio->bi_size; |
5961 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
5962 | &map_length, NULL, 0); | |
5963 | if (ret) { | |
64728bbb | 5964 | bio_put(orig_bio); |
e65e1535 MX |
5965 | return -EIO; |
5966 | } | |
5967 | ||
02f57c7a JB |
5968 | if (map_length >= orig_bio->bi_size) { |
5969 | bio = orig_bio; | |
5970 | goto submit; | |
5971 | } | |
5972 | ||
1ae39938 | 5973 | async_submit = 1; |
02f57c7a JB |
5974 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS); |
5975 | if (!bio) | |
5976 | return -ENOMEM; | |
5977 | bio->bi_private = dip; | |
5978 | bio->bi_end_io = btrfs_end_dio_bio; | |
5979 | atomic_inc(&dip->pending_bios); | |
5980 | ||
e65e1535 MX |
5981 | while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) { |
5982 | if (unlikely(map_length < submit_len + bvec->bv_len || | |
5983 | bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5984 | bvec->bv_offset) < bvec->bv_len)) { | |
5985 | /* | |
5986 | * inc the count before we submit the bio so | |
5987 | * we know the end IO handler won't happen before | |
5988 | * we inc the count. Otherwise, the dip might get freed | |
5989 | * before we're done setting it up | |
5990 | */ | |
5991 | atomic_inc(&dip->pending_bios); | |
5992 | ret = __btrfs_submit_dio_bio(bio, inode, rw, | |
5993 | file_offset, skip_sum, | |
1ae39938 | 5994 | csums, async_submit); |
e65e1535 MX |
5995 | if (ret) { |
5996 | bio_put(bio); | |
5997 | atomic_dec(&dip->pending_bios); | |
5998 | goto out_err; | |
5999 | } | |
6000 | ||
98bc3149 JB |
6001 | /* Write's use the ordered csums */ |
6002 | if (!write && !skip_sum) | |
e65e1535 MX |
6003 | csums = csums + nr_pages; |
6004 | start_sector += submit_len >> 9; | |
6005 | file_offset += submit_len; | |
6006 | ||
6007 | submit_len = 0; | |
6008 | nr_pages = 0; | |
6009 | ||
6010 | bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, | |
6011 | start_sector, GFP_NOFS); | |
6012 | if (!bio) | |
6013 | goto out_err; | |
6014 | bio->bi_private = dip; | |
6015 | bio->bi_end_io = btrfs_end_dio_bio; | |
6016 | ||
6017 | map_length = orig_bio->bi_size; | |
6018 | ret = btrfs_map_block(map_tree, READ, start_sector << 9, | |
6019 | &map_length, NULL, 0); | |
6020 | if (ret) { | |
6021 | bio_put(bio); | |
6022 | goto out_err; | |
6023 | } | |
6024 | } else { | |
6025 | submit_len += bvec->bv_len; | |
6026 | nr_pages ++; | |
6027 | bvec++; | |
6028 | } | |
6029 | } | |
6030 | ||
02f57c7a | 6031 | submit: |
e65e1535 | 6032 | ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum, |
1ae39938 | 6033 | csums, async_submit); |
e65e1535 MX |
6034 | if (!ret) |
6035 | return 0; | |
6036 | ||
6037 | bio_put(bio); | |
6038 | out_err: | |
6039 | dip->errors = 1; | |
6040 | /* | |
6041 | * before atomic variable goto zero, we must | |
6042 | * make sure dip->errors is perceived to be set. | |
6043 | */ | |
6044 | smp_mb__before_atomic_dec(); | |
6045 | if (atomic_dec_and_test(&dip->pending_bios)) | |
6046 | bio_io_error(dip->orig_bio); | |
6047 | ||
6048 | /* bio_end_io() will handle error, so we needn't return it */ | |
6049 | return 0; | |
6050 | } | |
6051 | ||
4b46fce2 JB |
6052 | static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode, |
6053 | loff_t file_offset) | |
6054 | { | |
6055 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
6056 | struct btrfs_dio_private *dip; | |
6057 | struct bio_vec *bvec = bio->bi_io_vec; | |
4b46fce2 | 6058 | int skip_sum; |
7b6d91da | 6059 | int write = rw & REQ_WRITE; |
4b46fce2 JB |
6060 | int ret = 0; |
6061 | ||
6062 | skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; | |
6063 | ||
6064 | dip = kmalloc(sizeof(*dip), GFP_NOFS); | |
6065 | if (!dip) { | |
6066 | ret = -ENOMEM; | |
6067 | goto free_ordered; | |
6068 | } | |
6069 | dip->csums = NULL; | |
6070 | ||
98bc3149 JB |
6071 | /* Write's use the ordered csum stuff, so we don't need dip->csums */ |
6072 | if (!write && !skip_sum) { | |
4b46fce2 JB |
6073 | dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS); |
6074 | if (!dip->csums) { | |
b4966b77 | 6075 | kfree(dip); |
4b46fce2 JB |
6076 | ret = -ENOMEM; |
6077 | goto free_ordered; | |
6078 | } | |
6079 | } | |
6080 | ||
6081 | dip->private = bio->bi_private; | |
6082 | dip->inode = inode; | |
6083 | dip->logical_offset = file_offset; | |
6084 | ||
4b46fce2 JB |
6085 | dip->bytes = 0; |
6086 | do { | |
6087 | dip->bytes += bvec->bv_len; | |
6088 | bvec++; | |
6089 | } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1)); | |
6090 | ||
46bfbb5c | 6091 | dip->disk_bytenr = (u64)bio->bi_sector << 9; |
4b46fce2 | 6092 | bio->bi_private = dip; |
e65e1535 MX |
6093 | dip->errors = 0; |
6094 | dip->orig_bio = bio; | |
6095 | atomic_set(&dip->pending_bios, 0); | |
4b46fce2 JB |
6096 | |
6097 | if (write) | |
6098 | bio->bi_end_io = btrfs_endio_direct_write; | |
6099 | else | |
6100 | bio->bi_end_io = btrfs_endio_direct_read; | |
6101 | ||
e65e1535 MX |
6102 | ret = btrfs_submit_direct_hook(rw, dip, skip_sum); |
6103 | if (!ret) | |
eaf25d93 | 6104 | return; |
4b46fce2 JB |
6105 | free_ordered: |
6106 | /* | |
6107 | * If this is a write, we need to clean up the reserved space and kill | |
6108 | * the ordered extent. | |
6109 | */ | |
6110 | if (write) { | |
6111 | struct btrfs_ordered_extent *ordered; | |
955256f2 | 6112 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
4b46fce2 JB |
6113 | if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) && |
6114 | !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) | |
6115 | btrfs_free_reserved_extent(root, ordered->start, | |
6116 | ordered->disk_len); | |
6117 | btrfs_put_ordered_extent(ordered); | |
6118 | btrfs_put_ordered_extent(ordered); | |
6119 | } | |
6120 | bio_endio(bio, ret); | |
6121 | } | |
6122 | ||
5a5f79b5 CM |
6123 | static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb, |
6124 | const struct iovec *iov, loff_t offset, | |
6125 | unsigned long nr_segs) | |
6126 | { | |
6127 | int seg; | |
a1b75f7d | 6128 | int i; |
5a5f79b5 CM |
6129 | size_t size; |
6130 | unsigned long addr; | |
6131 | unsigned blocksize_mask = root->sectorsize - 1; | |
6132 | ssize_t retval = -EINVAL; | |
6133 | loff_t end = offset; | |
6134 | ||
6135 | if (offset & blocksize_mask) | |
6136 | goto out; | |
6137 | ||
6138 | /* Check the memory alignment. Blocks cannot straddle pages */ | |
6139 | for (seg = 0; seg < nr_segs; seg++) { | |
6140 | addr = (unsigned long)iov[seg].iov_base; | |
6141 | size = iov[seg].iov_len; | |
6142 | end += size; | |
a1b75f7d | 6143 | if ((addr & blocksize_mask) || (size & blocksize_mask)) |
5a5f79b5 | 6144 | goto out; |
a1b75f7d JB |
6145 | |
6146 | /* If this is a write we don't need to check anymore */ | |
6147 | if (rw & WRITE) | |
6148 | continue; | |
6149 | ||
6150 | /* | |
6151 | * Check to make sure we don't have duplicate iov_base's in this | |
6152 | * iovec, if so return EINVAL, otherwise we'll get csum errors | |
6153 | * when reading back. | |
6154 | */ | |
6155 | for (i = seg + 1; i < nr_segs; i++) { | |
6156 | if (iov[seg].iov_base == iov[i].iov_base) | |
6157 | goto out; | |
6158 | } | |
5a5f79b5 CM |
6159 | } |
6160 | retval = 0; | |
6161 | out: | |
6162 | return retval; | |
6163 | } | |
16432985 CM |
6164 | static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb, |
6165 | const struct iovec *iov, loff_t offset, | |
6166 | unsigned long nr_segs) | |
6167 | { | |
4b46fce2 JB |
6168 | struct file *file = iocb->ki_filp; |
6169 | struct inode *inode = file->f_mapping->host; | |
6170 | struct btrfs_ordered_extent *ordered; | |
4845e44f | 6171 | struct extent_state *cached_state = NULL; |
4b46fce2 JB |
6172 | u64 lockstart, lockend; |
6173 | ssize_t ret; | |
4845e44f CM |
6174 | int writing = rw & WRITE; |
6175 | int write_bits = 0; | |
3f7c579c | 6176 | size_t count = iov_length(iov, nr_segs); |
4b46fce2 | 6177 | |
5a5f79b5 CM |
6178 | if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov, |
6179 | offset, nr_segs)) { | |
6180 | return 0; | |
6181 | } | |
6182 | ||
4b46fce2 | 6183 | lockstart = offset; |
3f7c579c CM |
6184 | lockend = offset + count - 1; |
6185 | ||
6186 | if (writing) { | |
6187 | ret = btrfs_delalloc_reserve_space(inode, count); | |
6188 | if (ret) | |
6189 | goto out; | |
6190 | } | |
4845e44f | 6191 | |
4b46fce2 | 6192 | while (1) { |
4845e44f CM |
6193 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6194 | 0, &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6195 | /* |
6196 | * We're concerned with the entire range that we're going to be | |
6197 | * doing DIO to, so we need to make sure theres no ordered | |
6198 | * extents in this range. | |
6199 | */ | |
6200 | ordered = btrfs_lookup_ordered_range(inode, lockstart, | |
6201 | lockend - lockstart + 1); | |
6202 | if (!ordered) | |
6203 | break; | |
4845e44f CM |
6204 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
6205 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6206 | btrfs_start_ordered_extent(inode, ordered, 1); |
6207 | btrfs_put_ordered_extent(ordered); | |
6208 | cond_resched(); | |
6209 | } | |
6210 | ||
4845e44f CM |
6211 | /* |
6212 | * we don't use btrfs_set_extent_delalloc because we don't want | |
6213 | * the dirty or uptodate bits | |
6214 | */ | |
6215 | if (writing) { | |
6216 | write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING; | |
6217 | ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend, | |
6218 | EXTENT_DELALLOC, 0, NULL, &cached_state, | |
6219 | GFP_NOFS); | |
6220 | if (ret) { | |
6221 | clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, | |
6222 | lockend, EXTENT_LOCKED | write_bits, | |
6223 | 1, 0, &cached_state, GFP_NOFS); | |
6224 | goto out; | |
6225 | } | |
6226 | } | |
6227 | ||
6228 | free_extent_state(cached_state); | |
6229 | cached_state = NULL; | |
6230 | ||
5a5f79b5 CM |
6231 | ret = __blockdev_direct_IO(rw, iocb, inode, |
6232 | BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev, | |
6233 | iov, offset, nr_segs, btrfs_get_blocks_direct, NULL, | |
6234 | btrfs_submit_direct, 0); | |
4b46fce2 JB |
6235 | |
6236 | if (ret < 0 && ret != -EIOCBQUEUED) { | |
4845e44f CM |
6237 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset, |
6238 | offset + iov_length(iov, nr_segs) - 1, | |
6239 | EXTENT_LOCKED | write_bits, 1, 0, | |
6240 | &cached_state, GFP_NOFS); | |
4b46fce2 JB |
6241 | } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) { |
6242 | /* | |
6243 | * We're falling back to buffered, unlock the section we didn't | |
6244 | * do IO on. | |
6245 | */ | |
4845e44f CM |
6246 | clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret, |
6247 | offset + iov_length(iov, nr_segs) - 1, | |
6248 | EXTENT_LOCKED | write_bits, 1, 0, | |
6249 | &cached_state, GFP_NOFS); | |
4b46fce2 | 6250 | } |
4845e44f CM |
6251 | out: |
6252 | free_extent_state(cached_state); | |
4b46fce2 | 6253 | return ret; |
16432985 CM |
6254 | } |
6255 | ||
1506fcc8 YS |
6256 | static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
6257 | __u64 start, __u64 len) | |
6258 | { | |
ec29ed5b | 6259 | return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap); |
1506fcc8 YS |
6260 | } |
6261 | ||
a52d9a80 | 6262 | int btrfs_readpage(struct file *file, struct page *page) |
9ebefb18 | 6263 | { |
d1310b2e CM |
6264 | struct extent_io_tree *tree; |
6265 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
8ddc7d9c | 6266 | return extent_read_full_page(tree, page, btrfs_get_extent, 0); |
9ebefb18 | 6267 | } |
1832a6d5 | 6268 | |
a52d9a80 | 6269 | static int btrfs_writepage(struct page *page, struct writeback_control *wbc) |
39279cc3 | 6270 | { |
d1310b2e | 6271 | struct extent_io_tree *tree; |
b888db2b CM |
6272 | |
6273 | ||
6274 | if (current->flags & PF_MEMALLOC) { | |
6275 | redirty_page_for_writepage(wbc, page); | |
6276 | unlock_page(page); | |
6277 | return 0; | |
6278 | } | |
d1310b2e | 6279 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
a52d9a80 | 6280 | return extent_write_full_page(tree, page, btrfs_get_extent, wbc); |
9ebefb18 CM |
6281 | } |
6282 | ||
f421950f CM |
6283 | int btrfs_writepages(struct address_space *mapping, |
6284 | struct writeback_control *wbc) | |
b293f02e | 6285 | { |
d1310b2e | 6286 | struct extent_io_tree *tree; |
771ed689 | 6287 | |
d1310b2e | 6288 | tree = &BTRFS_I(mapping->host)->io_tree; |
b293f02e CM |
6289 | return extent_writepages(tree, mapping, btrfs_get_extent, wbc); |
6290 | } | |
6291 | ||
3ab2fb5a CM |
6292 | static int |
6293 | btrfs_readpages(struct file *file, struct address_space *mapping, | |
6294 | struct list_head *pages, unsigned nr_pages) | |
6295 | { | |
d1310b2e CM |
6296 | struct extent_io_tree *tree; |
6297 | tree = &BTRFS_I(mapping->host)->io_tree; | |
3ab2fb5a CM |
6298 | return extent_readpages(tree, mapping, pages, nr_pages, |
6299 | btrfs_get_extent); | |
6300 | } | |
e6dcd2dc | 6301 | static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
9ebefb18 | 6302 | { |
d1310b2e CM |
6303 | struct extent_io_tree *tree; |
6304 | struct extent_map_tree *map; | |
a52d9a80 | 6305 | int ret; |
8c2383c3 | 6306 | |
d1310b2e CM |
6307 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
6308 | map = &BTRFS_I(page->mapping->host)->extent_tree; | |
70dec807 | 6309 | ret = try_release_extent_mapping(map, tree, page, gfp_flags); |
a52d9a80 CM |
6310 | if (ret == 1) { |
6311 | ClearPagePrivate(page); | |
6312 | set_page_private(page, 0); | |
6313 | page_cache_release(page); | |
39279cc3 | 6314 | } |
a52d9a80 | 6315 | return ret; |
39279cc3 CM |
6316 | } |
6317 | ||
e6dcd2dc CM |
6318 | static int btrfs_releasepage(struct page *page, gfp_t gfp_flags) |
6319 | { | |
98509cfc CM |
6320 | if (PageWriteback(page) || PageDirty(page)) |
6321 | return 0; | |
b335b003 | 6322 | return __btrfs_releasepage(page, gfp_flags & GFP_NOFS); |
e6dcd2dc CM |
6323 | } |
6324 | ||
a52d9a80 | 6325 | static void btrfs_invalidatepage(struct page *page, unsigned long offset) |
39279cc3 | 6326 | { |
d1310b2e | 6327 | struct extent_io_tree *tree; |
e6dcd2dc | 6328 | struct btrfs_ordered_extent *ordered; |
2ac55d41 | 6329 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6330 | u64 page_start = page_offset(page); |
6331 | u64 page_end = page_start + PAGE_CACHE_SIZE - 1; | |
39279cc3 | 6332 | |
8b62b72b CM |
6333 | |
6334 | /* | |
6335 | * we have the page locked, so new writeback can't start, | |
6336 | * and the dirty bit won't be cleared while we are here. | |
6337 | * | |
6338 | * Wait for IO on this page so that we can safely clear | |
6339 | * the PagePrivate2 bit and do ordered accounting | |
6340 | */ | |
e6dcd2dc | 6341 | wait_on_page_writeback(page); |
8b62b72b | 6342 | |
d1310b2e | 6343 | tree = &BTRFS_I(page->mapping->host)->io_tree; |
e6dcd2dc CM |
6344 | if (offset) { |
6345 | btrfs_releasepage(page, GFP_NOFS); | |
6346 | return; | |
6347 | } | |
2ac55d41 JB |
6348 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, |
6349 | GFP_NOFS); | |
e6dcd2dc CM |
6350 | ordered = btrfs_lookup_ordered_extent(page->mapping->host, |
6351 | page_offset(page)); | |
6352 | if (ordered) { | |
eb84ae03 CM |
6353 | /* |
6354 | * IO on this page will never be started, so we need | |
6355 | * to account for any ordered extents now | |
6356 | */ | |
e6dcd2dc CM |
6357 | clear_extent_bit(tree, page_start, page_end, |
6358 | EXTENT_DIRTY | EXTENT_DELALLOC | | |
32c00aff | 6359 | EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, |
2ac55d41 | 6360 | &cached_state, GFP_NOFS); |
8b62b72b CM |
6361 | /* |
6362 | * whoever cleared the private bit is responsible | |
6363 | * for the finish_ordered_io | |
6364 | */ | |
6365 | if (TestClearPagePrivate2(page)) { | |
6366 | btrfs_finish_ordered_io(page->mapping->host, | |
6367 | page_start, page_end); | |
6368 | } | |
e6dcd2dc | 6369 | btrfs_put_ordered_extent(ordered); |
2ac55d41 JB |
6370 | cached_state = NULL; |
6371 | lock_extent_bits(tree, page_start, page_end, 0, &cached_state, | |
6372 | GFP_NOFS); | |
e6dcd2dc CM |
6373 | } |
6374 | clear_extent_bit(tree, page_start, page_end, | |
32c00aff | 6375 | EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | |
2ac55d41 | 6376 | EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); |
e6dcd2dc CM |
6377 | __btrfs_releasepage(page, GFP_NOFS); |
6378 | ||
4a096752 | 6379 | ClearPageChecked(page); |
9ad6b7bc | 6380 | if (PagePrivate(page)) { |
9ad6b7bc CM |
6381 | ClearPagePrivate(page); |
6382 | set_page_private(page, 0); | |
6383 | page_cache_release(page); | |
6384 | } | |
39279cc3 CM |
6385 | } |
6386 | ||
9ebefb18 CM |
6387 | /* |
6388 | * btrfs_page_mkwrite() is not allowed to change the file size as it gets | |
6389 | * called from a page fault handler when a page is first dirtied. Hence we must | |
6390 | * be careful to check for EOF conditions here. We set the page up correctly | |
6391 | * for a written page which means we get ENOSPC checking when writing into | |
6392 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
6393 | * support these features. | |
6394 | * | |
6395 | * We are not allowed to take the i_mutex here so we have to play games to | |
6396 | * protect against truncate races as the page could now be beyond EOF. Because | |
6397 | * vmtruncate() writes the inode size before removing pages, once we have the | |
6398 | * page lock we can determine safely if the page is beyond EOF. If it is not | |
6399 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
6400 | * unlock the page. | |
6401 | */ | |
c2ec175c | 6402 | int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
9ebefb18 | 6403 | { |
c2ec175c | 6404 | struct page *page = vmf->page; |
6da6abae | 6405 | struct inode *inode = fdentry(vma->vm_file)->d_inode; |
1832a6d5 | 6406 | struct btrfs_root *root = BTRFS_I(inode)->root; |
e6dcd2dc CM |
6407 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; |
6408 | struct btrfs_ordered_extent *ordered; | |
2ac55d41 | 6409 | struct extent_state *cached_state = NULL; |
e6dcd2dc CM |
6410 | char *kaddr; |
6411 | unsigned long zero_start; | |
9ebefb18 | 6412 | loff_t size; |
1832a6d5 | 6413 | int ret; |
9998eb70 | 6414 | int reserved = 0; |
a52d9a80 | 6415 | u64 page_start; |
e6dcd2dc | 6416 | u64 page_end; |
9ebefb18 | 6417 | |
0ca1f7ce | 6418 | ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6419 | if (!ret) { |
22c44fe6 | 6420 | ret = btrfs_update_time(vma->vm_file); |
9998eb70 CM |
6421 | reserved = 1; |
6422 | } | |
56a76f82 NP |
6423 | if (ret) { |
6424 | if (ret == -ENOMEM) | |
6425 | ret = VM_FAULT_OOM; | |
6426 | else /* -ENOSPC, -EIO, etc */ | |
6427 | ret = VM_FAULT_SIGBUS; | |
9998eb70 CM |
6428 | if (reserved) |
6429 | goto out; | |
6430 | goto out_noreserve; | |
56a76f82 | 6431 | } |
1832a6d5 | 6432 | |
56a76f82 | 6433 | ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */ |
e6dcd2dc | 6434 | again: |
9ebefb18 | 6435 | lock_page(page); |
9ebefb18 | 6436 | size = i_size_read(inode); |
e6dcd2dc CM |
6437 | page_start = page_offset(page); |
6438 | page_end = page_start + PAGE_CACHE_SIZE - 1; | |
a52d9a80 | 6439 | |
9ebefb18 | 6440 | if ((page->mapping != inode->i_mapping) || |
e6dcd2dc | 6441 | (page_start >= size)) { |
9ebefb18 CM |
6442 | /* page got truncated out from underneath us */ |
6443 | goto out_unlock; | |
6444 | } | |
e6dcd2dc CM |
6445 | wait_on_page_writeback(page); |
6446 | ||
2ac55d41 JB |
6447 | lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state, |
6448 | GFP_NOFS); | |
e6dcd2dc CM |
6449 | set_page_extent_mapped(page); |
6450 | ||
eb84ae03 CM |
6451 | /* |
6452 | * we can't set the delalloc bits if there are pending ordered | |
6453 | * extents. Drop our locks and wait for them to finish | |
6454 | */ | |
e6dcd2dc CM |
6455 | ordered = btrfs_lookup_ordered_extent(inode, page_start); |
6456 | if (ordered) { | |
2ac55d41 JB |
6457 | unlock_extent_cached(io_tree, page_start, page_end, |
6458 | &cached_state, GFP_NOFS); | |
e6dcd2dc | 6459 | unlock_page(page); |
eb84ae03 | 6460 | btrfs_start_ordered_extent(inode, ordered, 1); |
e6dcd2dc CM |
6461 | btrfs_put_ordered_extent(ordered); |
6462 | goto again; | |
6463 | } | |
6464 | ||
fbf19087 JB |
6465 | /* |
6466 | * XXX - page_mkwrite gets called every time the page is dirtied, even | |
6467 | * if it was already dirty, so for space accounting reasons we need to | |
6468 | * clear any delalloc bits for the range we are fixing to save. There | |
6469 | * is probably a better way to do this, but for now keep consistent with | |
6470 | * prepare_pages in the normal write path. | |
6471 | */ | |
2ac55d41 | 6472 | clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, |
32c00aff | 6473 | EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, |
2ac55d41 | 6474 | 0, 0, &cached_state, GFP_NOFS); |
fbf19087 | 6475 | |
2ac55d41 JB |
6476 | ret = btrfs_set_extent_delalloc(inode, page_start, page_end, |
6477 | &cached_state); | |
9ed74f2d | 6478 | if (ret) { |
2ac55d41 JB |
6479 | unlock_extent_cached(io_tree, page_start, page_end, |
6480 | &cached_state, GFP_NOFS); | |
9ed74f2d JB |
6481 | ret = VM_FAULT_SIGBUS; |
6482 | goto out_unlock; | |
6483 | } | |
e6dcd2dc | 6484 | ret = 0; |
9ebefb18 CM |
6485 | |
6486 | /* page is wholly or partially inside EOF */ | |
a52d9a80 | 6487 | if (page_start + PAGE_CACHE_SIZE > size) |
e6dcd2dc | 6488 | zero_start = size & ~PAGE_CACHE_MASK; |
9ebefb18 | 6489 | else |
e6dcd2dc | 6490 | zero_start = PAGE_CACHE_SIZE; |
9ebefb18 | 6491 | |
e6dcd2dc CM |
6492 | if (zero_start != PAGE_CACHE_SIZE) { |
6493 | kaddr = kmap(page); | |
6494 | memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start); | |
6495 | flush_dcache_page(page); | |
6496 | kunmap(page); | |
6497 | } | |
247e743c | 6498 | ClearPageChecked(page); |
e6dcd2dc | 6499 | set_page_dirty(page); |
50a9b214 | 6500 | SetPageUptodate(page); |
5a3f23d5 | 6501 | |
257c62e1 CM |
6502 | BTRFS_I(inode)->last_trans = root->fs_info->generation; |
6503 | BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid; | |
6504 | ||
2ac55d41 | 6505 | unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS); |
9ebefb18 CM |
6506 | |
6507 | out_unlock: | |
50a9b214 CM |
6508 | if (!ret) |
6509 | return VM_FAULT_LOCKED; | |
9ebefb18 | 6510 | unlock_page(page); |
1832a6d5 | 6511 | out: |
ec39e180 | 6512 | btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); |
9998eb70 | 6513 | out_noreserve: |
9ebefb18 CM |
6514 | return ret; |
6515 | } | |
6516 | ||
a41ad394 | 6517 | static int btrfs_truncate(struct inode *inode) |
39279cc3 CM |
6518 | { |
6519 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
fcb80c2a | 6520 | struct btrfs_block_rsv *rsv; |
39279cc3 | 6521 | int ret; |
3893e33b | 6522 | int err = 0; |
39279cc3 | 6523 | struct btrfs_trans_handle *trans; |
d3c2fdcf | 6524 | unsigned long nr; |
dbe674a9 | 6525 | u64 mask = root->sectorsize - 1; |
07127184 | 6526 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
39279cc3 | 6527 | |
5d5e103a JB |
6528 | ret = btrfs_truncate_page(inode->i_mapping, inode->i_size); |
6529 | if (ret) | |
a41ad394 | 6530 | return ret; |
8082510e | 6531 | |
4a096752 | 6532 | btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); |
8082510e | 6533 | btrfs_ordered_update_i_size(inode, inode->i_size, NULL); |
39279cc3 | 6534 | |
fcb80c2a JB |
6535 | /* |
6536 | * Yes ladies and gentelment, this is indeed ugly. The fact is we have | |
6537 | * 3 things going on here | |
6538 | * | |
6539 | * 1) We need to reserve space for our orphan item and the space to | |
6540 | * delete our orphan item. Lord knows we don't want to have a dangling | |
6541 | * orphan item because we didn't reserve space to remove it. | |
6542 | * | |
6543 | * 2) We need to reserve space to update our inode. | |
6544 | * | |
6545 | * 3) We need to have something to cache all the space that is going to | |
6546 | * be free'd up by the truncate operation, but also have some slack | |
6547 | * space reserved in case it uses space during the truncate (thank you | |
6548 | * very much snapshotting). | |
6549 | * | |
6550 | * And we need these to all be seperate. The fact is we can use alot of | |
6551 | * space doing the truncate, and we have no earthly idea how much space | |
6552 | * we will use, so we need the truncate reservation to be seperate so it | |
6553 | * doesn't end up using space reserved for updating the inode or | |
6554 | * removing the orphan item. We also need to be able to stop the | |
6555 | * transaction and start a new one, which means we need to be able to | |
6556 | * update the inode several times, and we have no idea of knowing how | |
6557 | * many times that will be, so we can't just reserve 1 item for the | |
6558 | * entirety of the opration, so that has to be done seperately as well. | |
6559 | * Then there is the orphan item, which does indeed need to be held on | |
6560 | * to for the whole operation, and we need nobody to touch this reserved | |
6561 | * space except the orphan code. | |
6562 | * | |
6563 | * So that leaves us with | |
6564 | * | |
6565 | * 1) root->orphan_block_rsv - for the orphan deletion. | |
6566 | * 2) rsv - for the truncate reservation, which we will steal from the | |
6567 | * transaction reservation. | |
6568 | * 3) fs_info->trans_block_rsv - this will have 1 items worth left for | |
6569 | * updating the inode. | |
6570 | */ | |
6571 | rsv = btrfs_alloc_block_rsv(root); | |
6572 | if (!rsv) | |
6573 | return -ENOMEM; | |
4a338542 | 6574 | rsv->size = min_size; |
f0cd846e | 6575 | |
907cbceb | 6576 | /* |
07127184 | 6577 | * 1 for the truncate slack space |
907cbceb JB |
6578 | * 1 for the orphan item we're going to add |
6579 | * 1 for the orphan item deletion | |
6580 | * 1 for updating the inode. | |
6581 | */ | |
fcb80c2a JB |
6582 | trans = btrfs_start_transaction(root, 4); |
6583 | if (IS_ERR(trans)) { | |
6584 | err = PTR_ERR(trans); | |
6585 | goto out; | |
6586 | } | |
f0cd846e | 6587 | |
907cbceb JB |
6588 | /* Migrate the slack space for the truncate to our reserve */ |
6589 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, | |
6590 | min_size); | |
fcb80c2a | 6591 | BUG_ON(ret); |
f0cd846e JB |
6592 | |
6593 | ret = btrfs_orphan_add(trans, inode); | |
6594 | if (ret) { | |
6595 | btrfs_end_transaction(trans, root); | |
fcb80c2a | 6596 | goto out; |
f0cd846e JB |
6597 | } |
6598 | ||
5a3f23d5 CM |
6599 | /* |
6600 | * setattr is responsible for setting the ordered_data_close flag, | |
6601 | * but that is only tested during the last file release. That | |
6602 | * could happen well after the next commit, leaving a great big | |
6603 | * window where new writes may get lost if someone chooses to write | |
6604 | * to this file after truncating to zero | |
6605 | * | |
6606 | * The inode doesn't have any dirty data here, and so if we commit | |
6607 | * this is a noop. If someone immediately starts writing to the inode | |
6608 | * it is very likely we'll catch some of their writes in this | |
6609 | * transaction, and the commit will find this file on the ordered | |
6610 | * data list with good things to send down. | |
6611 | * | |
6612 | * This is a best effort solution, there is still a window where | |
6613 | * using truncate to replace the contents of the file will | |
6614 | * end up with a zero length file after a crash. | |
6615 | */ | |
6616 | if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close) | |
6617 | btrfs_add_ordered_operation(trans, root, inode); | |
6618 | ||
8082510e | 6619 | while (1) { |
36ba022a | 6620 | ret = btrfs_block_rsv_refill(root, rsv, min_size); |
907cbceb JB |
6621 | if (ret) { |
6622 | /* | |
6623 | * This can only happen with the original transaction we | |
6624 | * started above, every other time we shouldn't have a | |
6625 | * transaction started yet. | |
6626 | */ | |
6627 | if (ret == -EAGAIN) | |
6628 | goto end_trans; | |
6629 | err = ret; | |
6630 | break; | |
6631 | } | |
6632 | ||
d68fc57b | 6633 | if (!trans) { |
907cbceb JB |
6634 | /* Just need the 1 for updating the inode */ |
6635 | trans = btrfs_start_transaction(root, 1); | |
fcb80c2a | 6636 | if (IS_ERR(trans)) { |
7041ee97 JB |
6637 | ret = err = PTR_ERR(trans); |
6638 | trans = NULL; | |
6639 | break; | |
fcb80c2a | 6640 | } |
d68fc57b YZ |
6641 | } |
6642 | ||
907cbceb JB |
6643 | trans->block_rsv = rsv; |
6644 | ||
8082510e YZ |
6645 | ret = btrfs_truncate_inode_items(trans, root, inode, |
6646 | inode->i_size, | |
6647 | BTRFS_EXTENT_DATA_KEY); | |
3893e33b JB |
6648 | if (ret != -EAGAIN) { |
6649 | err = ret; | |
8082510e | 6650 | break; |
3893e33b | 6651 | } |
39279cc3 | 6652 | |
fcb80c2a | 6653 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
8082510e | 6654 | ret = btrfs_update_inode(trans, root, inode); |
3893e33b JB |
6655 | if (ret) { |
6656 | err = ret; | |
6657 | break; | |
6658 | } | |
907cbceb | 6659 | end_trans: |
8082510e YZ |
6660 | nr = trans->blocks_used; |
6661 | btrfs_end_transaction(trans, root); | |
d68fc57b | 6662 | trans = NULL; |
8082510e | 6663 | btrfs_btree_balance_dirty(root, nr); |
8082510e YZ |
6664 | } |
6665 | ||
6666 | if (ret == 0 && inode->i_nlink > 0) { | |
fcb80c2a | 6667 | trans->block_rsv = root->orphan_block_rsv; |
8082510e | 6668 | ret = btrfs_orphan_del(trans, inode); |
3893e33b JB |
6669 | if (ret) |
6670 | err = ret; | |
ded5db9d JB |
6671 | } else if (ret && inode->i_nlink > 0) { |
6672 | /* | |
6673 | * Failed to do the truncate, remove us from the in memory | |
6674 | * orphan list. | |
6675 | */ | |
6676 | ret = btrfs_orphan_del(NULL, inode); | |
8082510e YZ |
6677 | } |
6678 | ||
917c16b2 CM |
6679 | if (trans) { |
6680 | trans->block_rsv = &root->fs_info->trans_block_rsv; | |
6681 | ret = btrfs_update_inode(trans, root, inode); | |
6682 | if (ret && !err) | |
6683 | err = ret; | |
7b128766 | 6684 | |
917c16b2 | 6685 | nr = trans->blocks_used; |
7ad85bb7 | 6686 | ret = btrfs_end_transaction(trans, root); |
917c16b2 CM |
6687 | btrfs_btree_balance_dirty(root, nr); |
6688 | } | |
fcb80c2a JB |
6689 | |
6690 | out: | |
6691 | btrfs_free_block_rsv(root, rsv); | |
6692 | ||
3893e33b JB |
6693 | if (ret && !err) |
6694 | err = ret; | |
a41ad394 | 6695 | |
3893e33b | 6696 | return err; |
39279cc3 CM |
6697 | } |
6698 | ||
d352ac68 CM |
6699 | /* |
6700 | * create a new subvolume directory/inode (helper for the ioctl). | |
6701 | */ | |
d2fb3437 | 6702 | int btrfs_create_subvol_root(struct btrfs_trans_handle *trans, |
d82a6f1d | 6703 | struct btrfs_root *new_root, u64 new_dirid) |
39279cc3 | 6704 | { |
39279cc3 | 6705 | struct inode *inode; |
76dda93c | 6706 | int err; |
00e4e6b3 | 6707 | u64 index = 0; |
39279cc3 | 6708 | |
12fc9d09 FA |
6709 | inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, |
6710 | new_dirid, new_dirid, | |
6711 | S_IFDIR | (~current_umask() & S_IRWXUGO), | |
6712 | &index); | |
54aa1f4d | 6713 | if (IS_ERR(inode)) |
f46b5a66 | 6714 | return PTR_ERR(inode); |
39279cc3 CM |
6715 | inode->i_op = &btrfs_dir_inode_operations; |
6716 | inode->i_fop = &btrfs_dir_file_operations; | |
6717 | ||
bfe86848 | 6718 | set_nlink(inode, 1); |
dbe674a9 | 6719 | btrfs_i_size_write(inode, 0); |
3b96362c | 6720 | |
76dda93c YZ |
6721 | err = btrfs_update_inode(trans, new_root, inode); |
6722 | BUG_ON(err); | |
cb8e7090 | 6723 | |
76dda93c | 6724 | iput(inode); |
cb8e7090 | 6725 | return 0; |
39279cc3 CM |
6726 | } |
6727 | ||
39279cc3 CM |
6728 | struct inode *btrfs_alloc_inode(struct super_block *sb) |
6729 | { | |
6730 | struct btrfs_inode *ei; | |
2ead6ae7 | 6731 | struct inode *inode; |
39279cc3 CM |
6732 | |
6733 | ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS); | |
6734 | if (!ei) | |
6735 | return NULL; | |
2ead6ae7 YZ |
6736 | |
6737 | ei->root = NULL; | |
6738 | ei->space_info = NULL; | |
6739 | ei->generation = 0; | |
6740 | ei->sequence = 0; | |
15ee9bc7 | 6741 | ei->last_trans = 0; |
257c62e1 | 6742 | ei->last_sub_trans = 0; |
e02119d5 | 6743 | ei->logged_trans = 0; |
2ead6ae7 | 6744 | ei->delalloc_bytes = 0; |
2ead6ae7 YZ |
6745 | ei->disk_i_size = 0; |
6746 | ei->flags = 0; | |
7709cde3 | 6747 | ei->csum_bytes = 0; |
2ead6ae7 YZ |
6748 | ei->index_cnt = (u64)-1; |
6749 | ei->last_unlink_trans = 0; | |
6750 | ||
9e0baf60 JB |
6751 | spin_lock_init(&ei->lock); |
6752 | ei->outstanding_extents = 0; | |
6753 | ei->reserved_extents = 0; | |
2ead6ae7 YZ |
6754 | |
6755 | ei->ordered_data_close = 0; | |
d68fc57b | 6756 | ei->orphan_meta_reserved = 0; |
2ead6ae7 | 6757 | ei->dummy_inode = 0; |
4cb5300b | 6758 | ei->in_defrag = 0; |
7fd2ae21 | 6759 | ei->delalloc_meta_reserved = 0; |
261507a0 | 6760 | ei->force_compress = BTRFS_COMPRESS_NONE; |
2ead6ae7 | 6761 | |
16cdcec7 MX |
6762 | ei->delayed_node = NULL; |
6763 | ||
2ead6ae7 | 6764 | inode = &ei->vfs_inode; |
a8067e02 | 6765 | extent_map_tree_init(&ei->extent_tree); |
f993c883 DS |
6766 | extent_io_tree_init(&ei->io_tree, &inode->i_data); |
6767 | extent_io_tree_init(&ei->io_failure_tree, &inode->i_data); | |
2ead6ae7 | 6768 | mutex_init(&ei->log_mutex); |
f248679e | 6769 | mutex_init(&ei->delalloc_mutex); |
e6dcd2dc | 6770 | btrfs_ordered_inode_tree_init(&ei->ordered_tree); |
7b128766 | 6771 | INIT_LIST_HEAD(&ei->i_orphan); |
2ead6ae7 | 6772 | INIT_LIST_HEAD(&ei->delalloc_inodes); |
5a3f23d5 | 6773 | INIT_LIST_HEAD(&ei->ordered_operations); |
2ead6ae7 YZ |
6774 | RB_CLEAR_NODE(&ei->rb_node); |
6775 | ||
6776 | return inode; | |
39279cc3 CM |
6777 | } |
6778 | ||
fa0d7e3d NP |
6779 | static void btrfs_i_callback(struct rcu_head *head) |
6780 | { | |
6781 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
6782 | INIT_LIST_HEAD(&inode->i_dentry); | |
6783 | kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode)); | |
6784 | } | |
6785 | ||
39279cc3 CM |
6786 | void btrfs_destroy_inode(struct inode *inode) |
6787 | { | |
e6dcd2dc | 6788 | struct btrfs_ordered_extent *ordered; |
5a3f23d5 CM |
6789 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6790 | ||
39279cc3 CM |
6791 | WARN_ON(!list_empty(&inode->i_dentry)); |
6792 | WARN_ON(inode->i_data.nrpages); | |
9e0baf60 JB |
6793 | WARN_ON(BTRFS_I(inode)->outstanding_extents); |
6794 | WARN_ON(BTRFS_I(inode)->reserved_extents); | |
7709cde3 JB |
6795 | WARN_ON(BTRFS_I(inode)->delalloc_bytes); |
6796 | WARN_ON(BTRFS_I(inode)->csum_bytes); | |
39279cc3 | 6797 | |
a6dbd429 JB |
6798 | /* |
6799 | * This can happen where we create an inode, but somebody else also | |
6800 | * created the same inode and we need to destroy the one we already | |
6801 | * created. | |
6802 | */ | |
6803 | if (!root) | |
6804 | goto free; | |
6805 | ||
5a3f23d5 CM |
6806 | /* |
6807 | * Make sure we're properly removed from the ordered operation | |
6808 | * lists. | |
6809 | */ | |
6810 | smp_mb(); | |
6811 | if (!list_empty(&BTRFS_I(inode)->ordered_operations)) { | |
6812 | spin_lock(&root->fs_info->ordered_extent_lock); | |
6813 | list_del_init(&BTRFS_I(inode)->ordered_operations); | |
6814 | spin_unlock(&root->fs_info->ordered_extent_lock); | |
6815 | } | |
6816 | ||
d68fc57b | 6817 | spin_lock(&root->orphan_lock); |
7b128766 | 6818 | if (!list_empty(&BTRFS_I(inode)->i_orphan)) { |
33345d01 LZ |
6819 | printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n", |
6820 | (unsigned long long)btrfs_ino(inode)); | |
8082510e | 6821 | list_del_init(&BTRFS_I(inode)->i_orphan); |
7b128766 | 6822 | } |
d68fc57b | 6823 | spin_unlock(&root->orphan_lock); |
7b128766 | 6824 | |
d397712b | 6825 | while (1) { |
e6dcd2dc CM |
6826 | ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); |
6827 | if (!ordered) | |
6828 | break; | |
6829 | else { | |
d397712b CM |
6830 | printk(KERN_ERR "btrfs found ordered " |
6831 | "extent %llu %llu on inode cleanup\n", | |
6832 | (unsigned long long)ordered->file_offset, | |
6833 | (unsigned long long)ordered->len); | |
e6dcd2dc CM |
6834 | btrfs_remove_ordered_extent(inode, ordered); |
6835 | btrfs_put_ordered_extent(ordered); | |
6836 | btrfs_put_ordered_extent(ordered); | |
6837 | } | |
6838 | } | |
5d4f98a2 | 6839 | inode_tree_del(inode); |
5b21f2ed | 6840 | btrfs_drop_extent_cache(inode, 0, (u64)-1, 0); |
a6dbd429 | 6841 | free: |
16cdcec7 | 6842 | btrfs_remove_delayed_node(inode); |
fa0d7e3d | 6843 | call_rcu(&inode->i_rcu, btrfs_i_callback); |
39279cc3 CM |
6844 | } |
6845 | ||
45321ac5 | 6846 | int btrfs_drop_inode(struct inode *inode) |
76dda93c YZ |
6847 | { |
6848 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
45321ac5 | 6849 | |
0af3d00b | 6850 | if (btrfs_root_refs(&root->root_item) == 0 && |
2cf8572d | 6851 | !btrfs_is_free_space_inode(root, inode)) |
45321ac5 | 6852 | return 1; |
76dda93c | 6853 | else |
45321ac5 | 6854 | return generic_drop_inode(inode); |
76dda93c YZ |
6855 | } |
6856 | ||
0ee0fda0 | 6857 | static void init_once(void *foo) |
39279cc3 CM |
6858 | { |
6859 | struct btrfs_inode *ei = (struct btrfs_inode *) foo; | |
6860 | ||
6861 | inode_init_once(&ei->vfs_inode); | |
6862 | } | |
6863 | ||
6864 | void btrfs_destroy_cachep(void) | |
6865 | { | |
6866 | if (btrfs_inode_cachep) | |
6867 | kmem_cache_destroy(btrfs_inode_cachep); | |
6868 | if (btrfs_trans_handle_cachep) | |
6869 | kmem_cache_destroy(btrfs_trans_handle_cachep); | |
6870 | if (btrfs_transaction_cachep) | |
6871 | kmem_cache_destroy(btrfs_transaction_cachep); | |
39279cc3 CM |
6872 | if (btrfs_path_cachep) |
6873 | kmem_cache_destroy(btrfs_path_cachep); | |
dc89e982 JB |
6874 | if (btrfs_free_space_cachep) |
6875 | kmem_cache_destroy(btrfs_free_space_cachep); | |
39279cc3 CM |
6876 | } |
6877 | ||
6878 | int btrfs_init_cachep(void) | |
6879 | { | |
9601e3f6 CH |
6880 | btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache", |
6881 | sizeof(struct btrfs_inode), 0, | |
6882 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once); | |
39279cc3 CM |
6883 | if (!btrfs_inode_cachep) |
6884 | goto fail; | |
9601e3f6 CH |
6885 | |
6886 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache", | |
6887 | sizeof(struct btrfs_trans_handle), 0, | |
6888 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6889 | if (!btrfs_trans_handle_cachep) |
6890 | goto fail; | |
9601e3f6 CH |
6891 | |
6892 | btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache", | |
6893 | sizeof(struct btrfs_transaction), 0, | |
6894 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6895 | if (!btrfs_transaction_cachep) |
6896 | goto fail; | |
9601e3f6 CH |
6897 | |
6898 | btrfs_path_cachep = kmem_cache_create("btrfs_path_cache", | |
6899 | sizeof(struct btrfs_path), 0, | |
6900 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
39279cc3 CM |
6901 | if (!btrfs_path_cachep) |
6902 | goto fail; | |
9601e3f6 | 6903 | |
dc89e982 JB |
6904 | btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache", |
6905 | sizeof(struct btrfs_free_space), 0, | |
6906 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL); | |
6907 | if (!btrfs_free_space_cachep) | |
6908 | goto fail; | |
6909 | ||
39279cc3 CM |
6910 | return 0; |
6911 | fail: | |
6912 | btrfs_destroy_cachep(); | |
6913 | return -ENOMEM; | |
6914 | } | |
6915 | ||
6916 | static int btrfs_getattr(struct vfsmount *mnt, | |
6917 | struct dentry *dentry, struct kstat *stat) | |
6918 | { | |
6919 | struct inode *inode = dentry->d_inode; | |
fadc0d8b DS |
6920 | u32 blocksize = inode->i_sb->s_blocksize; |
6921 | ||
39279cc3 | 6922 | generic_fillattr(inode, stat); |
0ee5dc67 | 6923 | stat->dev = BTRFS_I(inode)->root->anon_dev; |
d6667462 | 6924 | stat->blksize = PAGE_CACHE_SIZE; |
fadc0d8b DS |
6925 | stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) + |
6926 | ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9; | |
39279cc3 CM |
6927 | return 0; |
6928 | } | |
6929 | ||
75e7cb7f LB |
6930 | /* |
6931 | * If a file is moved, it will inherit the cow and compression flags of the new | |
6932 | * directory. | |
6933 | */ | |
6934 | static void fixup_inode_flags(struct inode *dir, struct inode *inode) | |
6935 | { | |
6936 | struct btrfs_inode *b_dir = BTRFS_I(dir); | |
6937 | struct btrfs_inode *b_inode = BTRFS_I(inode); | |
6938 | ||
6939 | if (b_dir->flags & BTRFS_INODE_NODATACOW) | |
6940 | b_inode->flags |= BTRFS_INODE_NODATACOW; | |
6941 | else | |
6942 | b_inode->flags &= ~BTRFS_INODE_NODATACOW; | |
6943 | ||
6944 | if (b_dir->flags & BTRFS_INODE_COMPRESS) | |
6945 | b_inode->flags |= BTRFS_INODE_COMPRESS; | |
6946 | else | |
6947 | b_inode->flags &= ~BTRFS_INODE_COMPRESS; | |
6948 | } | |
6949 | ||
d397712b CM |
6950 | static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry, |
6951 | struct inode *new_dir, struct dentry *new_dentry) | |
39279cc3 CM |
6952 | { |
6953 | struct btrfs_trans_handle *trans; | |
6954 | struct btrfs_root *root = BTRFS_I(old_dir)->root; | |
4df27c4d | 6955 | struct btrfs_root *dest = BTRFS_I(new_dir)->root; |
39279cc3 CM |
6956 | struct inode *new_inode = new_dentry->d_inode; |
6957 | struct inode *old_inode = old_dentry->d_inode; | |
6958 | struct timespec ctime = CURRENT_TIME; | |
00e4e6b3 | 6959 | u64 index = 0; |
4df27c4d | 6960 | u64 root_objectid; |
39279cc3 | 6961 | int ret; |
33345d01 | 6962 | u64 old_ino = btrfs_ino(old_inode); |
39279cc3 | 6963 | |
33345d01 | 6964 | if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) |
f679a840 YZ |
6965 | return -EPERM; |
6966 | ||
4df27c4d | 6967 | /* we only allow rename subvolume link between subvolumes */ |
33345d01 | 6968 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest) |
3394e160 CM |
6969 | return -EXDEV; |
6970 | ||
33345d01 LZ |
6971 | if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID || |
6972 | (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID)) | |
39279cc3 | 6973 | return -ENOTEMPTY; |
5f39d397 | 6974 | |
4df27c4d YZ |
6975 | if (S_ISDIR(old_inode->i_mode) && new_inode && |
6976 | new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) | |
6977 | return -ENOTEMPTY; | |
5a3f23d5 CM |
6978 | /* |
6979 | * we're using rename to replace one file with another. | |
6980 | * and the replacement file is large. Start IO on it now so | |
6981 | * we don't add too much work to the end of the transaction | |
6982 | */ | |
4baf8c92 | 6983 | if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size && |
5a3f23d5 CM |
6984 | old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT) |
6985 | filemap_flush(old_inode->i_mapping); | |
6986 | ||
76dda93c | 6987 | /* close the racy window with snapshot create/destroy ioctl */ |
33345d01 | 6988 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 6989 | down_read(&root->fs_info->subvol_sem); |
a22285a6 YZ |
6990 | /* |
6991 | * We want to reserve the absolute worst case amount of items. So if | |
6992 | * both inodes are subvols and we need to unlink them then that would | |
6993 | * require 4 item modifications, but if they are both normal inodes it | |
6994 | * would require 5 item modifications, so we'll assume their normal | |
6995 | * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items | |
6996 | * should cover the worst case number of items we'll modify. | |
6997 | */ | |
6998 | trans = btrfs_start_transaction(root, 20); | |
b44c59a8 JL |
6999 | if (IS_ERR(trans)) { |
7000 | ret = PTR_ERR(trans); | |
7001 | goto out_notrans; | |
7002 | } | |
76dda93c | 7003 | |
4df27c4d YZ |
7004 | if (dest != root) |
7005 | btrfs_record_root_in_trans(trans, dest); | |
5f39d397 | 7006 | |
a5719521 YZ |
7007 | ret = btrfs_set_inode_index(new_dir, &index); |
7008 | if (ret) | |
7009 | goto out_fail; | |
5a3f23d5 | 7010 | |
33345d01 | 7011 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7012 | /* force full log commit if subvolume involved. */ |
7013 | root->fs_info->last_trans_log_full_commit = trans->transid; | |
7014 | } else { | |
a5719521 YZ |
7015 | ret = btrfs_insert_inode_ref(trans, dest, |
7016 | new_dentry->d_name.name, | |
7017 | new_dentry->d_name.len, | |
33345d01 LZ |
7018 | old_ino, |
7019 | btrfs_ino(new_dir), index); | |
a5719521 YZ |
7020 | if (ret) |
7021 | goto out_fail; | |
4df27c4d YZ |
7022 | /* |
7023 | * this is an ugly little race, but the rename is required | |
7024 | * to make sure that if we crash, the inode is either at the | |
7025 | * old name or the new one. pinning the log transaction lets | |
7026 | * us make sure we don't allow a log commit to come in after | |
7027 | * we unlink the name but before we add the new name back in. | |
7028 | */ | |
7029 | btrfs_pin_log_trans(root); | |
7030 | } | |
5a3f23d5 CM |
7031 | /* |
7032 | * make sure the inode gets flushed if it is replacing | |
7033 | * something. | |
7034 | */ | |
33345d01 | 7035 | if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode)) |
5a3f23d5 | 7036 | btrfs_add_ordered_operation(trans, root, old_inode); |
5a3f23d5 | 7037 | |
39279cc3 CM |
7038 | old_dir->i_ctime = old_dir->i_mtime = ctime; |
7039 | new_dir->i_ctime = new_dir->i_mtime = ctime; | |
7040 | old_inode->i_ctime = ctime; | |
5f39d397 | 7041 | |
12fcfd22 CM |
7042 | if (old_dentry->d_parent != new_dentry->d_parent) |
7043 | btrfs_record_unlink_dir(trans, old_dir, old_inode, 1); | |
7044 | ||
33345d01 | 7045 | if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) { |
4df27c4d YZ |
7046 | root_objectid = BTRFS_I(old_inode)->root->root_key.objectid; |
7047 | ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid, | |
7048 | old_dentry->d_name.name, | |
7049 | old_dentry->d_name.len); | |
7050 | } else { | |
92986796 AV |
7051 | ret = __btrfs_unlink_inode(trans, root, old_dir, |
7052 | old_dentry->d_inode, | |
7053 | old_dentry->d_name.name, | |
7054 | old_dentry->d_name.len); | |
7055 | if (!ret) | |
7056 | ret = btrfs_update_inode(trans, root, old_inode); | |
4df27c4d YZ |
7057 | } |
7058 | BUG_ON(ret); | |
39279cc3 CM |
7059 | |
7060 | if (new_inode) { | |
7061 | new_inode->i_ctime = CURRENT_TIME; | |
33345d01 | 7062 | if (unlikely(btrfs_ino(new_inode) == |
4df27c4d YZ |
7063 | BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) { |
7064 | root_objectid = BTRFS_I(new_inode)->location.objectid; | |
7065 | ret = btrfs_unlink_subvol(trans, dest, new_dir, | |
7066 | root_objectid, | |
7067 | new_dentry->d_name.name, | |
7068 | new_dentry->d_name.len); | |
7069 | BUG_ON(new_inode->i_nlink == 0); | |
7070 | } else { | |
7071 | ret = btrfs_unlink_inode(trans, dest, new_dir, | |
7072 | new_dentry->d_inode, | |
7073 | new_dentry->d_name.name, | |
7074 | new_dentry->d_name.len); | |
7075 | } | |
7076 | BUG_ON(ret); | |
7b128766 | 7077 | if (new_inode->i_nlink == 0) { |
e02119d5 | 7078 | ret = btrfs_orphan_add(trans, new_dentry->d_inode); |
4df27c4d | 7079 | BUG_ON(ret); |
7b128766 | 7080 | } |
39279cc3 | 7081 | } |
aec7477b | 7082 | |
75e7cb7f LB |
7083 | fixup_inode_flags(new_dir, old_inode); |
7084 | ||
4df27c4d YZ |
7085 | ret = btrfs_add_link(trans, new_dir, old_inode, |
7086 | new_dentry->d_name.name, | |
a5719521 | 7087 | new_dentry->d_name.len, 0, index); |
4df27c4d | 7088 | BUG_ON(ret); |
39279cc3 | 7089 | |
33345d01 | 7090 | if (old_ino != BTRFS_FIRST_FREE_OBJECTID) { |
10d9f309 | 7091 | struct dentry *parent = new_dentry->d_parent; |
6a912213 | 7092 | btrfs_log_new_name(trans, old_inode, old_dir, parent); |
4df27c4d YZ |
7093 | btrfs_end_log_trans(root); |
7094 | } | |
39279cc3 | 7095 | out_fail: |
7ad85bb7 | 7096 | btrfs_end_transaction(trans, root); |
b44c59a8 | 7097 | out_notrans: |
33345d01 | 7098 | if (old_ino == BTRFS_FIRST_FREE_OBJECTID) |
76dda93c | 7099 | up_read(&root->fs_info->subvol_sem); |
9ed74f2d | 7100 | |
39279cc3 CM |
7101 | return ret; |
7102 | } | |
7103 | ||
d352ac68 CM |
7104 | /* |
7105 | * some fairly slow code that needs optimization. This walks the list | |
7106 | * of all the inodes with pending delalloc and forces them to disk. | |
7107 | */ | |
24bbcf04 | 7108 | int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput) |
ea8c2819 CM |
7109 | { |
7110 | struct list_head *head = &root->fs_info->delalloc_inodes; | |
7111 | struct btrfs_inode *binode; | |
5b21f2ed | 7112 | struct inode *inode; |
ea8c2819 | 7113 | |
c146afad YZ |
7114 | if (root->fs_info->sb->s_flags & MS_RDONLY) |
7115 | return -EROFS; | |
7116 | ||
75eff68e | 7117 | spin_lock(&root->fs_info->delalloc_lock); |
d397712b | 7118 | while (!list_empty(head)) { |
ea8c2819 CM |
7119 | binode = list_entry(head->next, struct btrfs_inode, |
7120 | delalloc_inodes); | |
5b21f2ed ZY |
7121 | inode = igrab(&binode->vfs_inode); |
7122 | if (!inode) | |
7123 | list_del_init(&binode->delalloc_inodes); | |
75eff68e | 7124 | spin_unlock(&root->fs_info->delalloc_lock); |
5b21f2ed | 7125 | if (inode) { |
8c8bee1d | 7126 | filemap_flush(inode->i_mapping); |
24bbcf04 YZ |
7127 | if (delay_iput) |
7128 | btrfs_add_delayed_iput(inode); | |
7129 | else | |
7130 | iput(inode); | |
5b21f2ed ZY |
7131 | } |
7132 | cond_resched(); | |
75eff68e | 7133 | spin_lock(&root->fs_info->delalloc_lock); |
ea8c2819 | 7134 | } |
75eff68e | 7135 | spin_unlock(&root->fs_info->delalloc_lock); |
8c8bee1d CM |
7136 | |
7137 | /* the filemap_flush will queue IO into the worker threads, but | |
7138 | * we have to make sure the IO is actually started and that | |
7139 | * ordered extents get created before we return | |
7140 | */ | |
7141 | atomic_inc(&root->fs_info->async_submit_draining); | |
d397712b | 7142 | while (atomic_read(&root->fs_info->nr_async_submits) || |
771ed689 | 7143 | atomic_read(&root->fs_info->async_delalloc_pages)) { |
8c8bee1d | 7144 | wait_event(root->fs_info->async_submit_wait, |
771ed689 CM |
7145 | (atomic_read(&root->fs_info->nr_async_submits) == 0 && |
7146 | atomic_read(&root->fs_info->async_delalloc_pages) == 0)); | |
8c8bee1d CM |
7147 | } |
7148 | atomic_dec(&root->fs_info->async_submit_draining); | |
ea8c2819 CM |
7149 | return 0; |
7150 | } | |
7151 | ||
39279cc3 CM |
7152 | static int btrfs_symlink(struct inode *dir, struct dentry *dentry, |
7153 | const char *symname) | |
7154 | { | |
7155 | struct btrfs_trans_handle *trans; | |
7156 | struct btrfs_root *root = BTRFS_I(dir)->root; | |
7157 | struct btrfs_path *path; | |
7158 | struct btrfs_key key; | |
1832a6d5 | 7159 | struct inode *inode = NULL; |
39279cc3 CM |
7160 | int err; |
7161 | int drop_inode = 0; | |
7162 | u64 objectid; | |
00e4e6b3 | 7163 | u64 index = 0 ; |
39279cc3 CM |
7164 | int name_len; |
7165 | int datasize; | |
5f39d397 | 7166 | unsigned long ptr; |
39279cc3 | 7167 | struct btrfs_file_extent_item *ei; |
5f39d397 | 7168 | struct extent_buffer *leaf; |
1832a6d5 | 7169 | unsigned long nr = 0; |
39279cc3 CM |
7170 | |
7171 | name_len = strlen(symname) + 1; | |
7172 | if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root)) | |
7173 | return -ENAMETOOLONG; | |
1832a6d5 | 7174 | |
9ed74f2d JB |
7175 | /* |
7176 | * 2 items for inode item and ref | |
7177 | * 2 items for dir items | |
7178 | * 1 item for xattr if selinux is on | |
7179 | */ | |
a22285a6 YZ |
7180 | trans = btrfs_start_transaction(root, 5); |
7181 | if (IS_ERR(trans)) | |
7182 | return PTR_ERR(trans); | |
1832a6d5 | 7183 | |
581bb050 LZ |
7184 | err = btrfs_find_free_ino(root, &objectid); |
7185 | if (err) | |
7186 | goto out_unlock; | |
7187 | ||
aec7477b | 7188 | inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name, |
33345d01 | 7189 | dentry->d_name.len, btrfs_ino(dir), objectid, |
d82a6f1d | 7190 | S_IFLNK|S_IRWXUGO, &index); |
7cf96da3 TI |
7191 | if (IS_ERR(inode)) { |
7192 | err = PTR_ERR(inode); | |
39279cc3 | 7193 | goto out_unlock; |
7cf96da3 | 7194 | } |
39279cc3 | 7195 | |
2a7dba39 | 7196 | err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name); |
33268eaf JB |
7197 | if (err) { |
7198 | drop_inode = 1; | |
7199 | goto out_unlock; | |
7200 | } | |
7201 | ||
ad19db71 CS |
7202 | /* |
7203 | * If the active LSM wants to access the inode during | |
7204 | * d_instantiate it needs these. Smack checks to see | |
7205 | * if the filesystem supports xattrs by looking at the | |
7206 | * ops vector. | |
7207 | */ | |
7208 | inode->i_fop = &btrfs_file_operations; | |
7209 | inode->i_op = &btrfs_file_inode_operations; | |
7210 | ||
a1b075d2 | 7211 | err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index); |
39279cc3 CM |
7212 | if (err) |
7213 | drop_inode = 1; | |
7214 | else { | |
7215 | inode->i_mapping->a_ops = &btrfs_aops; | |
04160088 | 7216 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d1310b2e | 7217 | BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; |
39279cc3 | 7218 | } |
39279cc3 CM |
7219 | if (drop_inode) |
7220 | goto out_unlock; | |
7221 | ||
7222 | path = btrfs_alloc_path(); | |
d8926bb3 MF |
7223 | if (!path) { |
7224 | err = -ENOMEM; | |
7225 | drop_inode = 1; | |
7226 | goto out_unlock; | |
7227 | } | |
33345d01 | 7228 | key.objectid = btrfs_ino(inode); |
39279cc3 | 7229 | key.offset = 0; |
39279cc3 CM |
7230 | btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY); |
7231 | datasize = btrfs_file_extent_calc_inline_size(name_len); | |
7232 | err = btrfs_insert_empty_item(trans, root, path, &key, | |
7233 | datasize); | |
54aa1f4d CM |
7234 | if (err) { |
7235 | drop_inode = 1; | |
b0839166 | 7236 | btrfs_free_path(path); |
54aa1f4d CM |
7237 | goto out_unlock; |
7238 | } | |
5f39d397 CM |
7239 | leaf = path->nodes[0]; |
7240 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
7241 | struct btrfs_file_extent_item); | |
7242 | btrfs_set_file_extent_generation(leaf, ei, trans->transid); | |
7243 | btrfs_set_file_extent_type(leaf, ei, | |
39279cc3 | 7244 | BTRFS_FILE_EXTENT_INLINE); |
c8b97818 CM |
7245 | btrfs_set_file_extent_encryption(leaf, ei, 0); |
7246 | btrfs_set_file_extent_compression(leaf, ei, 0); | |
7247 | btrfs_set_file_extent_other_encoding(leaf, ei, 0); | |
7248 | btrfs_set_file_extent_ram_bytes(leaf, ei, name_len); | |
7249 | ||
39279cc3 | 7250 | ptr = btrfs_file_extent_inline_start(ei); |
5f39d397 CM |
7251 | write_extent_buffer(leaf, symname, ptr, name_len); |
7252 | btrfs_mark_buffer_dirty(leaf); | |
39279cc3 | 7253 | btrfs_free_path(path); |
5f39d397 | 7254 | |
39279cc3 CM |
7255 | inode->i_op = &btrfs_symlink_inode_operations; |
7256 | inode->i_mapping->a_ops = &btrfs_symlink_aops; | |
04160088 | 7257 | inode->i_mapping->backing_dev_info = &root->fs_info->bdi; |
d899e052 | 7258 | inode_set_bytes(inode, name_len); |
dbe674a9 | 7259 | btrfs_i_size_write(inode, name_len - 1); |
54aa1f4d CM |
7260 | err = btrfs_update_inode(trans, root, inode); |
7261 | if (err) | |
7262 | drop_inode = 1; | |
39279cc3 CM |
7263 | |
7264 | out_unlock: | |
08c422c2 AV |
7265 | if (!err) |
7266 | d_instantiate(dentry, inode); | |
d3c2fdcf | 7267 | nr = trans->blocks_used; |
7ad85bb7 | 7268 | btrfs_end_transaction(trans, root); |
39279cc3 CM |
7269 | if (drop_inode) { |
7270 | inode_dec_link_count(inode); | |
7271 | iput(inode); | |
7272 | } | |
d3c2fdcf | 7273 | btrfs_btree_balance_dirty(root, nr); |
39279cc3 CM |
7274 | return err; |
7275 | } | |
16432985 | 7276 | |
0af3d00b JB |
7277 | static int __btrfs_prealloc_file_range(struct inode *inode, int mode, |
7278 | u64 start, u64 num_bytes, u64 min_size, | |
7279 | loff_t actual_len, u64 *alloc_hint, | |
7280 | struct btrfs_trans_handle *trans) | |
d899e052 | 7281 | { |
d899e052 YZ |
7282 | struct btrfs_root *root = BTRFS_I(inode)->root; |
7283 | struct btrfs_key ins; | |
d899e052 | 7284 | u64 cur_offset = start; |
55a61d1d | 7285 | u64 i_size; |
d899e052 | 7286 | int ret = 0; |
0af3d00b | 7287 | bool own_trans = true; |
d899e052 | 7288 | |
0af3d00b JB |
7289 | if (trans) |
7290 | own_trans = false; | |
d899e052 | 7291 | while (num_bytes > 0) { |
0af3d00b JB |
7292 | if (own_trans) { |
7293 | trans = btrfs_start_transaction(root, 3); | |
7294 | if (IS_ERR(trans)) { | |
7295 | ret = PTR_ERR(trans); | |
7296 | break; | |
7297 | } | |
5a303d5d YZ |
7298 | } |
7299 | ||
efa56464 YZ |
7300 | ret = btrfs_reserve_extent(trans, root, num_bytes, min_size, |
7301 | 0, *alloc_hint, (u64)-1, &ins, 1); | |
5a303d5d | 7302 | if (ret) { |
0af3d00b JB |
7303 | if (own_trans) |
7304 | btrfs_end_transaction(trans, root); | |
a22285a6 | 7305 | break; |
d899e052 | 7306 | } |
5a303d5d | 7307 | |
d899e052 YZ |
7308 | ret = insert_reserved_file_extent(trans, inode, |
7309 | cur_offset, ins.objectid, | |
7310 | ins.offset, ins.offset, | |
920bbbfb | 7311 | ins.offset, 0, 0, 0, |
d899e052 YZ |
7312 | BTRFS_FILE_EXTENT_PREALLOC); |
7313 | BUG_ON(ret); | |
a1ed835e CM |
7314 | btrfs_drop_extent_cache(inode, cur_offset, |
7315 | cur_offset + ins.offset -1, 0); | |
5a303d5d | 7316 | |
d899e052 YZ |
7317 | num_bytes -= ins.offset; |
7318 | cur_offset += ins.offset; | |
efa56464 | 7319 | *alloc_hint = ins.objectid + ins.offset; |
5a303d5d | 7320 | |
d899e052 | 7321 | inode->i_ctime = CURRENT_TIME; |
6cbff00f | 7322 | BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC; |
d899e052 | 7323 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
efa56464 YZ |
7324 | (actual_len > inode->i_size) && |
7325 | (cur_offset > inode->i_size)) { | |
d1ea6a61 | 7326 | if (cur_offset > actual_len) |
55a61d1d | 7327 | i_size = actual_len; |
d1ea6a61 | 7328 | else |
55a61d1d JB |
7329 | i_size = cur_offset; |
7330 | i_size_write(inode, i_size); | |
7331 | btrfs_ordered_update_i_size(inode, i_size, NULL); | |
5a303d5d YZ |
7332 | } |
7333 | ||
d899e052 YZ |
7334 | ret = btrfs_update_inode(trans, root, inode); |
7335 | BUG_ON(ret); | |
d899e052 | 7336 | |
0af3d00b JB |
7337 | if (own_trans) |
7338 | btrfs_end_transaction(trans, root); | |
5a303d5d | 7339 | } |
d899e052 YZ |
7340 | return ret; |
7341 | } | |
7342 | ||
0af3d00b JB |
7343 | int btrfs_prealloc_file_range(struct inode *inode, int mode, |
7344 | u64 start, u64 num_bytes, u64 min_size, | |
7345 | loff_t actual_len, u64 *alloc_hint) | |
7346 | { | |
7347 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7348 | min_size, actual_len, alloc_hint, | |
7349 | NULL); | |
7350 | } | |
7351 | ||
7352 | int btrfs_prealloc_file_range_trans(struct inode *inode, | |
7353 | struct btrfs_trans_handle *trans, int mode, | |
7354 | u64 start, u64 num_bytes, u64 min_size, | |
7355 | loff_t actual_len, u64 *alloc_hint) | |
7356 | { | |
7357 | return __btrfs_prealloc_file_range(inode, mode, start, num_bytes, | |
7358 | min_size, actual_len, alloc_hint, trans); | |
7359 | } | |
7360 | ||
e6dcd2dc CM |
7361 | static int btrfs_set_page_dirty(struct page *page) |
7362 | { | |
e6dcd2dc CM |
7363 | return __set_page_dirty_nobuffers(page); |
7364 | } | |
7365 | ||
10556cb2 | 7366 | static int btrfs_permission(struct inode *inode, int mask) |
fdebe2bd | 7367 | { |
b83cc969 | 7368 | struct btrfs_root *root = BTRFS_I(inode)->root; |
cb6db4e5 | 7369 | umode_t mode = inode->i_mode; |
b83cc969 | 7370 | |
cb6db4e5 JM |
7371 | if (mask & MAY_WRITE && |
7372 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
7373 | if (btrfs_root_readonly(root)) | |
7374 | return -EROFS; | |
7375 | if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) | |
7376 | return -EACCES; | |
7377 | } | |
2830ba7f | 7378 | return generic_permission(inode, mask); |
fdebe2bd | 7379 | } |
39279cc3 | 7380 | |
6e1d5dcc | 7381 | static const struct inode_operations btrfs_dir_inode_operations = { |
3394e160 | 7382 | .getattr = btrfs_getattr, |
39279cc3 CM |
7383 | .lookup = btrfs_lookup, |
7384 | .create = btrfs_create, | |
7385 | .unlink = btrfs_unlink, | |
7386 | .link = btrfs_link, | |
7387 | .mkdir = btrfs_mkdir, | |
7388 | .rmdir = btrfs_rmdir, | |
7389 | .rename = btrfs_rename, | |
7390 | .symlink = btrfs_symlink, | |
7391 | .setattr = btrfs_setattr, | |
618e21d5 | 7392 | .mknod = btrfs_mknod, |
95819c05 CH |
7393 | .setxattr = btrfs_setxattr, |
7394 | .getxattr = btrfs_getxattr, | |
5103e947 | 7395 | .listxattr = btrfs_listxattr, |
95819c05 | 7396 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7397 | .permission = btrfs_permission, |
4e34e719 | 7398 | .get_acl = btrfs_get_acl, |
39279cc3 | 7399 | }; |
6e1d5dcc | 7400 | static const struct inode_operations btrfs_dir_ro_inode_operations = { |
39279cc3 | 7401 | .lookup = btrfs_lookup, |
fdebe2bd | 7402 | .permission = btrfs_permission, |
4e34e719 | 7403 | .get_acl = btrfs_get_acl, |
39279cc3 | 7404 | }; |
76dda93c | 7405 | |
828c0950 | 7406 | static const struct file_operations btrfs_dir_file_operations = { |
39279cc3 CM |
7407 | .llseek = generic_file_llseek, |
7408 | .read = generic_read_dir, | |
cbdf5a24 | 7409 | .readdir = btrfs_real_readdir, |
34287aa3 | 7410 | .unlocked_ioctl = btrfs_ioctl, |
39279cc3 | 7411 | #ifdef CONFIG_COMPAT |
34287aa3 | 7412 | .compat_ioctl = btrfs_ioctl, |
39279cc3 | 7413 | #endif |
6bf13c0c | 7414 | .release = btrfs_release_file, |
e02119d5 | 7415 | .fsync = btrfs_sync_file, |
39279cc3 CM |
7416 | }; |
7417 | ||
d1310b2e | 7418 | static struct extent_io_ops btrfs_extent_io_ops = { |
07157aac | 7419 | .fill_delalloc = run_delalloc_range, |
065631f6 | 7420 | .submit_bio_hook = btrfs_submit_bio_hook, |
239b14b3 | 7421 | .merge_bio_hook = btrfs_merge_bio_hook, |
07157aac | 7422 | .readpage_end_io_hook = btrfs_readpage_end_io_hook, |
e6dcd2dc | 7423 | .writepage_end_io_hook = btrfs_writepage_end_io_hook, |
247e743c | 7424 | .writepage_start_hook = btrfs_writepage_start_hook, |
b0c68f8b CM |
7425 | .set_bit_hook = btrfs_set_bit_hook, |
7426 | .clear_bit_hook = btrfs_clear_bit_hook, | |
9ed74f2d JB |
7427 | .merge_extent_hook = btrfs_merge_extent_hook, |
7428 | .split_extent_hook = btrfs_split_extent_hook, | |
07157aac CM |
7429 | }; |
7430 | ||
35054394 CM |
7431 | /* |
7432 | * btrfs doesn't support the bmap operation because swapfiles | |
7433 | * use bmap to make a mapping of extents in the file. They assume | |
7434 | * these extents won't change over the life of the file and they | |
7435 | * use the bmap result to do IO directly to the drive. | |
7436 | * | |
7437 | * the btrfs bmap call would return logical addresses that aren't | |
7438 | * suitable for IO and they also will change frequently as COW | |
7439 | * operations happen. So, swapfile + btrfs == corruption. | |
7440 | * | |
7441 | * For now we're avoiding this by dropping bmap. | |
7442 | */ | |
7f09410b | 7443 | static const struct address_space_operations btrfs_aops = { |
39279cc3 CM |
7444 | .readpage = btrfs_readpage, |
7445 | .writepage = btrfs_writepage, | |
b293f02e | 7446 | .writepages = btrfs_writepages, |
3ab2fb5a | 7447 | .readpages = btrfs_readpages, |
16432985 | 7448 | .direct_IO = btrfs_direct_IO, |
a52d9a80 CM |
7449 | .invalidatepage = btrfs_invalidatepage, |
7450 | .releasepage = btrfs_releasepage, | |
e6dcd2dc | 7451 | .set_page_dirty = btrfs_set_page_dirty, |
465fdd97 | 7452 | .error_remove_page = generic_error_remove_page, |
39279cc3 CM |
7453 | }; |
7454 | ||
7f09410b | 7455 | static const struct address_space_operations btrfs_symlink_aops = { |
39279cc3 CM |
7456 | .readpage = btrfs_readpage, |
7457 | .writepage = btrfs_writepage, | |
2bf5a725 CM |
7458 | .invalidatepage = btrfs_invalidatepage, |
7459 | .releasepage = btrfs_releasepage, | |
39279cc3 CM |
7460 | }; |
7461 | ||
6e1d5dcc | 7462 | static const struct inode_operations btrfs_file_inode_operations = { |
39279cc3 CM |
7463 | .getattr = btrfs_getattr, |
7464 | .setattr = btrfs_setattr, | |
95819c05 CH |
7465 | .setxattr = btrfs_setxattr, |
7466 | .getxattr = btrfs_getxattr, | |
5103e947 | 7467 | .listxattr = btrfs_listxattr, |
95819c05 | 7468 | .removexattr = btrfs_removexattr, |
fdebe2bd | 7469 | .permission = btrfs_permission, |
1506fcc8 | 7470 | .fiemap = btrfs_fiemap, |
4e34e719 | 7471 | .get_acl = btrfs_get_acl, |
39279cc3 | 7472 | }; |
6e1d5dcc | 7473 | static const struct inode_operations btrfs_special_inode_operations = { |
618e21d5 JB |
7474 | .getattr = btrfs_getattr, |
7475 | .setattr = btrfs_setattr, | |
fdebe2bd | 7476 | .permission = btrfs_permission, |
95819c05 CH |
7477 | .setxattr = btrfs_setxattr, |
7478 | .getxattr = btrfs_getxattr, | |
33268eaf | 7479 | .listxattr = btrfs_listxattr, |
95819c05 | 7480 | .removexattr = btrfs_removexattr, |
4e34e719 | 7481 | .get_acl = btrfs_get_acl, |
618e21d5 | 7482 | }; |
6e1d5dcc | 7483 | static const struct inode_operations btrfs_symlink_inode_operations = { |
39279cc3 CM |
7484 | .readlink = generic_readlink, |
7485 | .follow_link = page_follow_link_light, | |
7486 | .put_link = page_put_link, | |
f209561a | 7487 | .getattr = btrfs_getattr, |
22c44fe6 | 7488 | .setattr = btrfs_setattr, |
fdebe2bd | 7489 | .permission = btrfs_permission, |
0279b4cd JO |
7490 | .setxattr = btrfs_setxattr, |
7491 | .getxattr = btrfs_getxattr, | |
7492 | .listxattr = btrfs_listxattr, | |
7493 | .removexattr = btrfs_removexattr, | |
4e34e719 | 7494 | .get_acl = btrfs_get_acl, |
39279cc3 | 7495 | }; |
76dda93c | 7496 | |
82d339d9 | 7497 | const struct dentry_operations btrfs_dentry_operations = { |
76dda93c | 7498 | .d_delete = btrfs_dentry_delete, |
b4aff1f8 | 7499 | .d_release = btrfs_dentry_release, |
76dda93c | 7500 | }; |